1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
21 #include "print-tree.h"
25 #include "free-space-cache.h"
26 #include "free-space-tree.h"
30 #include "ref-verify.h"
32 #undef SCRAMBLE_DELAYED_REFS
35 * control flags for do_chunk_alloc's force field
36 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
37 * if we really need one.
39 * CHUNK_ALLOC_LIMITED means to only try and allocate one
40 * if we have very few chunks already allocated. This is
41 * used as part of the clustering code to help make sure
42 * we have a good pool of storage to cluster in, without
43 * filling the FS with empty chunks
45 * CHUNK_ALLOC_FORCE means it must try to allocate one
49 CHUNK_ALLOC_NO_FORCE = 0,
50 CHUNK_ALLOC_LIMITED = 1,
51 CHUNK_ALLOC_FORCE = 2,
54 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
55 struct btrfs_fs_info *fs_info,
56 struct btrfs_delayed_ref_node *node, u64 parent,
57 u64 root_objectid, u64 owner_objectid,
58 u64 owner_offset, int refs_to_drop,
59 struct btrfs_delayed_extent_op *extra_op);
60 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
61 struct extent_buffer *leaf,
62 struct btrfs_extent_item *ei);
63 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
64 struct btrfs_fs_info *fs_info,
65 u64 parent, u64 root_objectid,
66 u64 flags, u64 owner, u64 offset,
67 struct btrfs_key *ins, int ref_mod);
68 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
69 struct btrfs_fs_info *fs_info,
70 u64 parent, u64 root_objectid,
71 u64 flags, struct btrfs_disk_key *key,
72 int level, struct btrfs_key *ins);
73 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
74 struct btrfs_fs_info *fs_info, u64 flags,
76 static int find_next_key(struct btrfs_path *path, int level,
77 struct btrfs_key *key);
78 static void dump_space_info(struct btrfs_fs_info *fs_info,
79 struct btrfs_space_info *info, u64 bytes,
80 int dump_block_groups);
81 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
83 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
84 struct btrfs_space_info *space_info,
86 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
87 struct btrfs_space_info *space_info,
91 block_group_cache_done(struct btrfs_block_group_cache *cache)
94 return cache->cached == BTRFS_CACHE_FINISHED ||
95 cache->cached == BTRFS_CACHE_ERROR;
98 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
100 return (cache->flags & bits) == bits;
103 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
105 atomic_inc(&cache->count);
108 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
110 if (atomic_dec_and_test(&cache->count)) {
111 WARN_ON(cache->pinned > 0);
112 WARN_ON(cache->reserved > 0);
115 * If not empty, someone is still holding mutex of
116 * full_stripe_lock, which can only be released by caller.
117 * And it will definitely cause use-after-free when caller
118 * tries to release full stripe lock.
120 * No better way to resolve, but only to warn.
122 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
123 kfree(cache->free_space_ctl);
129 * this adds the block group to the fs_info rb tree for the block group
132 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
133 struct btrfs_block_group_cache *block_group)
136 struct rb_node *parent = NULL;
137 struct btrfs_block_group_cache *cache;
139 spin_lock(&info->block_group_cache_lock);
140 p = &info->block_group_cache_tree.rb_node;
144 cache = rb_entry(parent, struct btrfs_block_group_cache,
146 if (block_group->key.objectid < cache->key.objectid) {
148 } else if (block_group->key.objectid > cache->key.objectid) {
151 spin_unlock(&info->block_group_cache_lock);
156 rb_link_node(&block_group->cache_node, parent, p);
157 rb_insert_color(&block_group->cache_node,
158 &info->block_group_cache_tree);
160 if (info->first_logical_byte > block_group->key.objectid)
161 info->first_logical_byte = block_group->key.objectid;
163 spin_unlock(&info->block_group_cache_lock);
169 * This will return the block group at or after bytenr if contains is 0, else
170 * it will return the block group that contains the bytenr
172 static struct btrfs_block_group_cache *
173 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
176 struct btrfs_block_group_cache *cache, *ret = NULL;
180 spin_lock(&info->block_group_cache_lock);
181 n = info->block_group_cache_tree.rb_node;
184 cache = rb_entry(n, struct btrfs_block_group_cache,
186 end = cache->key.objectid + cache->key.offset - 1;
187 start = cache->key.objectid;
189 if (bytenr < start) {
190 if (!contains && (!ret || start < ret->key.objectid))
193 } else if (bytenr > start) {
194 if (contains && bytenr <= end) {
205 btrfs_get_block_group(ret);
206 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
207 info->first_logical_byte = ret->key.objectid;
209 spin_unlock(&info->block_group_cache_lock);
214 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
215 u64 start, u64 num_bytes)
217 u64 end = start + num_bytes - 1;
218 set_extent_bits(&fs_info->freed_extents[0],
219 start, end, EXTENT_UPTODATE);
220 set_extent_bits(&fs_info->freed_extents[1],
221 start, end, EXTENT_UPTODATE);
225 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
226 struct btrfs_block_group_cache *cache)
230 start = cache->key.objectid;
231 end = start + cache->key.offset - 1;
233 clear_extent_bits(&fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE);
235 clear_extent_bits(&fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE);
239 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
240 struct btrfs_block_group_cache *cache)
247 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
248 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
249 cache->bytes_super += stripe_len;
250 ret = add_excluded_extent(fs_info, cache->key.objectid,
256 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
257 bytenr = btrfs_sb_offset(i);
258 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
259 bytenr, 0, &logical, &nr, &stripe_len);
266 if (logical[nr] > cache->key.objectid +
270 if (logical[nr] + stripe_len <= cache->key.objectid)
274 if (start < cache->key.objectid) {
275 start = cache->key.objectid;
276 len = (logical[nr] + stripe_len) - start;
278 len = min_t(u64, stripe_len,
279 cache->key.objectid +
280 cache->key.offset - start);
283 cache->bytes_super += len;
284 ret = add_excluded_extent(fs_info, start, len);
296 static struct btrfs_caching_control *
297 get_caching_control(struct btrfs_block_group_cache *cache)
299 struct btrfs_caching_control *ctl;
301 spin_lock(&cache->lock);
302 if (!cache->caching_ctl) {
303 spin_unlock(&cache->lock);
307 ctl = cache->caching_ctl;
308 refcount_inc(&ctl->count);
309 spin_unlock(&cache->lock);
313 static void put_caching_control(struct btrfs_caching_control *ctl)
315 if (refcount_dec_and_test(&ctl->count))
319 #ifdef CONFIG_BTRFS_DEBUG
320 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
322 struct btrfs_fs_info *fs_info = block_group->fs_info;
323 u64 start = block_group->key.objectid;
324 u64 len = block_group->key.offset;
325 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
326 fs_info->nodesize : fs_info->sectorsize;
327 u64 step = chunk << 1;
329 while (len > chunk) {
330 btrfs_remove_free_space(block_group, start, chunk);
341 * this is only called by cache_block_group, since we could have freed extents
342 * we need to check the pinned_extents for any extents that can't be used yet
343 * since their free space will be released as soon as the transaction commits.
345 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
346 struct btrfs_fs_info *info, u64 start, u64 end)
348 u64 extent_start, extent_end, size, total_added = 0;
351 while (start < end) {
352 ret = find_first_extent_bit(info->pinned_extents, start,
353 &extent_start, &extent_end,
354 EXTENT_DIRTY | EXTENT_UPTODATE,
359 if (extent_start <= start) {
360 start = extent_end + 1;
361 } else if (extent_start > start && extent_start < end) {
362 size = extent_start - start;
364 ret = btrfs_add_free_space(block_group, start,
366 BUG_ON(ret); /* -ENOMEM or logic error */
367 start = extent_end + 1;
376 ret = btrfs_add_free_space(block_group, start, size);
377 BUG_ON(ret); /* -ENOMEM or logic error */
383 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
385 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
386 struct btrfs_fs_info *fs_info = block_group->fs_info;
387 struct btrfs_root *extent_root = fs_info->extent_root;
388 struct btrfs_path *path;
389 struct extent_buffer *leaf;
390 struct btrfs_key key;
397 path = btrfs_alloc_path();
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
403 #ifdef CONFIG_BTRFS_DEBUG
405 * If we're fragmenting we don't want to make anybody think we can
406 * allocate from this block group until we've had a chance to fragment
409 if (btrfs_should_fragment_free_space(block_group))
413 * We don't want to deadlock with somebody trying to allocate a new
414 * extent for the extent root while also trying to search the extent
415 * root to add free space. So we skip locking and search the commit
416 * root, since its read-only
418 path->skip_locking = 1;
419 path->search_commit_root = 1;
420 path->reada = READA_FORWARD;
424 key.type = BTRFS_EXTENT_ITEM_KEY;
427 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
431 leaf = path->nodes[0];
432 nritems = btrfs_header_nritems(leaf);
435 if (btrfs_fs_closing(fs_info) > 1) {
440 if (path->slots[0] < nritems) {
441 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
443 ret = find_next_key(path, 0, &key);
447 if (need_resched() ||
448 rwsem_is_contended(&fs_info->commit_root_sem)) {
450 caching_ctl->progress = last;
451 btrfs_release_path(path);
452 up_read(&fs_info->commit_root_sem);
453 mutex_unlock(&caching_ctl->mutex);
455 mutex_lock(&caching_ctl->mutex);
456 down_read(&fs_info->commit_root_sem);
460 ret = btrfs_next_leaf(extent_root, path);
465 leaf = path->nodes[0];
466 nritems = btrfs_header_nritems(leaf);
470 if (key.objectid < last) {
473 key.type = BTRFS_EXTENT_ITEM_KEY;
476 caching_ctl->progress = last;
477 btrfs_release_path(path);
481 if (key.objectid < block_group->key.objectid) {
486 if (key.objectid >= block_group->key.objectid +
487 block_group->key.offset)
490 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
491 key.type == BTRFS_METADATA_ITEM_KEY) {
492 total_found += add_new_free_space(block_group,
495 if (key.type == BTRFS_METADATA_ITEM_KEY)
496 last = key.objectid +
499 last = key.objectid + key.offset;
501 if (total_found > CACHING_CTL_WAKE_UP) {
504 wake_up(&caching_ctl->wait);
511 total_found += add_new_free_space(block_group, fs_info, last,
512 block_group->key.objectid +
513 block_group->key.offset);
514 caching_ctl->progress = (u64)-1;
517 btrfs_free_path(path);
521 static noinline void caching_thread(struct btrfs_work *work)
523 struct btrfs_block_group_cache *block_group;
524 struct btrfs_fs_info *fs_info;
525 struct btrfs_caching_control *caching_ctl;
528 caching_ctl = container_of(work, struct btrfs_caching_control, work);
529 block_group = caching_ctl->block_group;
530 fs_info = block_group->fs_info;
532 mutex_lock(&caching_ctl->mutex);
533 down_read(&fs_info->commit_root_sem);
535 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
536 ret = load_free_space_tree(caching_ctl);
538 ret = load_extent_tree_free(caching_ctl);
540 spin_lock(&block_group->lock);
541 block_group->caching_ctl = NULL;
542 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
543 spin_unlock(&block_group->lock);
545 #ifdef CONFIG_BTRFS_DEBUG
546 if (btrfs_should_fragment_free_space(block_group)) {
549 spin_lock(&block_group->space_info->lock);
550 spin_lock(&block_group->lock);
551 bytes_used = block_group->key.offset -
552 btrfs_block_group_used(&block_group->item);
553 block_group->space_info->bytes_used += bytes_used >> 1;
554 spin_unlock(&block_group->lock);
555 spin_unlock(&block_group->space_info->lock);
556 fragment_free_space(block_group);
560 caching_ctl->progress = (u64)-1;
562 up_read(&fs_info->commit_root_sem);
563 free_excluded_extents(fs_info, block_group);
564 mutex_unlock(&caching_ctl->mutex);
566 wake_up(&caching_ctl->wait);
568 put_caching_control(caching_ctl);
569 btrfs_put_block_group(block_group);
572 static int cache_block_group(struct btrfs_block_group_cache *cache,
576 struct btrfs_fs_info *fs_info = cache->fs_info;
577 struct btrfs_caching_control *caching_ctl;
580 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
584 INIT_LIST_HEAD(&caching_ctl->list);
585 mutex_init(&caching_ctl->mutex);
586 init_waitqueue_head(&caching_ctl->wait);
587 caching_ctl->block_group = cache;
588 caching_ctl->progress = cache->key.objectid;
589 refcount_set(&caching_ctl->count, 1);
590 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
591 caching_thread, NULL, NULL);
593 spin_lock(&cache->lock);
595 * This should be a rare occasion, but this could happen I think in the
596 * case where one thread starts to load the space cache info, and then
597 * some other thread starts a transaction commit which tries to do an
598 * allocation while the other thread is still loading the space cache
599 * info. The previous loop should have kept us from choosing this block
600 * group, but if we've moved to the state where we will wait on caching
601 * block groups we need to first check if we're doing a fast load here,
602 * so we can wait for it to finish, otherwise we could end up allocating
603 * from a block group who's cache gets evicted for one reason or
606 while (cache->cached == BTRFS_CACHE_FAST) {
607 struct btrfs_caching_control *ctl;
609 ctl = cache->caching_ctl;
610 refcount_inc(&ctl->count);
611 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
612 spin_unlock(&cache->lock);
616 finish_wait(&ctl->wait, &wait);
617 put_caching_control(ctl);
618 spin_lock(&cache->lock);
621 if (cache->cached != BTRFS_CACHE_NO) {
622 spin_unlock(&cache->lock);
626 WARN_ON(cache->caching_ctl);
627 cache->caching_ctl = caching_ctl;
628 cache->cached = BTRFS_CACHE_FAST;
629 spin_unlock(&cache->lock);
631 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
632 mutex_lock(&caching_ctl->mutex);
633 ret = load_free_space_cache(fs_info, cache);
635 spin_lock(&cache->lock);
637 cache->caching_ctl = NULL;
638 cache->cached = BTRFS_CACHE_FINISHED;
639 cache->last_byte_to_unpin = (u64)-1;
640 caching_ctl->progress = (u64)-1;
642 if (load_cache_only) {
643 cache->caching_ctl = NULL;
644 cache->cached = BTRFS_CACHE_NO;
646 cache->cached = BTRFS_CACHE_STARTED;
647 cache->has_caching_ctl = 1;
650 spin_unlock(&cache->lock);
651 #ifdef CONFIG_BTRFS_DEBUG
653 btrfs_should_fragment_free_space(cache)) {
656 spin_lock(&cache->space_info->lock);
657 spin_lock(&cache->lock);
658 bytes_used = cache->key.offset -
659 btrfs_block_group_used(&cache->item);
660 cache->space_info->bytes_used += bytes_used >> 1;
661 spin_unlock(&cache->lock);
662 spin_unlock(&cache->space_info->lock);
663 fragment_free_space(cache);
666 mutex_unlock(&caching_ctl->mutex);
668 wake_up(&caching_ctl->wait);
670 put_caching_control(caching_ctl);
671 free_excluded_extents(fs_info, cache);
676 * We're either using the free space tree or no caching at all.
677 * Set cached to the appropriate value and wakeup any waiters.
679 spin_lock(&cache->lock);
680 if (load_cache_only) {
681 cache->caching_ctl = NULL;
682 cache->cached = BTRFS_CACHE_NO;
684 cache->cached = BTRFS_CACHE_STARTED;
685 cache->has_caching_ctl = 1;
687 spin_unlock(&cache->lock);
688 wake_up(&caching_ctl->wait);
691 if (load_cache_only) {
692 put_caching_control(caching_ctl);
696 down_write(&fs_info->commit_root_sem);
697 refcount_inc(&caching_ctl->count);
698 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
699 up_write(&fs_info->commit_root_sem);
701 btrfs_get_block_group(cache);
703 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
709 * return the block group that starts at or after bytenr
711 static struct btrfs_block_group_cache *
712 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
714 return block_group_cache_tree_search(info, bytenr, 0);
718 * return the block group that contains the given bytenr
720 struct btrfs_block_group_cache *btrfs_lookup_block_group(
721 struct btrfs_fs_info *info,
724 return block_group_cache_tree_search(info, bytenr, 1);
727 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
730 struct list_head *head = &info->space_info;
731 struct btrfs_space_info *found;
733 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
736 list_for_each_entry_rcu(found, head, list) {
737 if (found->flags & flags) {
746 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
747 u64 owner, u64 root_objectid)
749 struct btrfs_space_info *space_info;
752 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
753 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
754 flags = BTRFS_BLOCK_GROUP_SYSTEM;
756 flags = BTRFS_BLOCK_GROUP_METADATA;
758 flags = BTRFS_BLOCK_GROUP_DATA;
761 space_info = __find_space_info(fs_info, flags);
763 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
767 * after adding space to the filesystem, we need to clear the full flags
768 * on all the space infos.
770 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
772 struct list_head *head = &info->space_info;
773 struct btrfs_space_info *found;
776 list_for_each_entry_rcu(found, head, list)
781 /* simple helper to search for an existing data extent at a given offset */
782 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
785 struct btrfs_key key;
786 struct btrfs_path *path;
788 path = btrfs_alloc_path();
792 key.objectid = start;
794 key.type = BTRFS_EXTENT_ITEM_KEY;
795 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
796 btrfs_free_path(path);
801 * helper function to lookup reference count and flags of a tree block.
803 * the head node for delayed ref is used to store the sum of all the
804 * reference count modifications queued up in the rbtree. the head
805 * node may also store the extent flags to set. This way you can check
806 * to see what the reference count and extent flags would be if all of
807 * the delayed refs are not processed.
809 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
810 struct btrfs_fs_info *fs_info, u64 bytenr,
811 u64 offset, int metadata, u64 *refs, u64 *flags)
813 struct btrfs_delayed_ref_head *head;
814 struct btrfs_delayed_ref_root *delayed_refs;
815 struct btrfs_path *path;
816 struct btrfs_extent_item *ei;
817 struct extent_buffer *leaf;
818 struct btrfs_key key;
825 * If we don't have skinny metadata, don't bother doing anything
828 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
829 offset = fs_info->nodesize;
833 path = btrfs_alloc_path();
838 path->skip_locking = 1;
839 path->search_commit_root = 1;
843 key.objectid = bytenr;
846 key.type = BTRFS_METADATA_ITEM_KEY;
848 key.type = BTRFS_EXTENT_ITEM_KEY;
850 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
854 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
855 if (path->slots[0]) {
857 btrfs_item_key_to_cpu(path->nodes[0], &key,
859 if (key.objectid == bytenr &&
860 key.type == BTRFS_EXTENT_ITEM_KEY &&
861 key.offset == fs_info->nodesize)
867 leaf = path->nodes[0];
868 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
869 if (item_size >= sizeof(*ei)) {
870 ei = btrfs_item_ptr(leaf, path->slots[0],
871 struct btrfs_extent_item);
872 num_refs = btrfs_extent_refs(leaf, ei);
873 extent_flags = btrfs_extent_flags(leaf, ei);
875 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
876 struct btrfs_extent_item_v0 *ei0;
877 BUG_ON(item_size != sizeof(*ei0));
878 ei0 = btrfs_item_ptr(leaf, path->slots[0],
879 struct btrfs_extent_item_v0);
880 num_refs = btrfs_extent_refs_v0(leaf, ei0);
881 /* FIXME: this isn't correct for data */
882 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
887 BUG_ON(num_refs == 0);
897 delayed_refs = &trans->transaction->delayed_refs;
898 spin_lock(&delayed_refs->lock);
899 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
901 if (!mutex_trylock(&head->mutex)) {
902 refcount_inc(&head->refs);
903 spin_unlock(&delayed_refs->lock);
905 btrfs_release_path(path);
908 * Mutex was contended, block until it's released and try
911 mutex_lock(&head->mutex);
912 mutex_unlock(&head->mutex);
913 btrfs_put_delayed_ref_head(head);
916 spin_lock(&head->lock);
917 if (head->extent_op && head->extent_op->update_flags)
918 extent_flags |= head->extent_op->flags_to_set;
920 BUG_ON(num_refs == 0);
922 num_refs += head->ref_mod;
923 spin_unlock(&head->lock);
924 mutex_unlock(&head->mutex);
926 spin_unlock(&delayed_refs->lock);
928 WARN_ON(num_refs == 0);
932 *flags = extent_flags;
934 btrfs_free_path(path);
939 * Back reference rules. Back refs have three main goals:
941 * 1) differentiate between all holders of references to an extent so that
942 * when a reference is dropped we can make sure it was a valid reference
943 * before freeing the extent.
945 * 2) Provide enough information to quickly find the holders of an extent
946 * if we notice a given block is corrupted or bad.
948 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
949 * maintenance. This is actually the same as #2, but with a slightly
950 * different use case.
952 * There are two kinds of back refs. The implicit back refs is optimized
953 * for pointers in non-shared tree blocks. For a given pointer in a block,
954 * back refs of this kind provide information about the block's owner tree
955 * and the pointer's key. These information allow us to find the block by
956 * b-tree searching. The full back refs is for pointers in tree blocks not
957 * referenced by their owner trees. The location of tree block is recorded
958 * in the back refs. Actually the full back refs is generic, and can be
959 * used in all cases the implicit back refs is used. The major shortcoming
960 * of the full back refs is its overhead. Every time a tree block gets
961 * COWed, we have to update back refs entry for all pointers in it.
963 * For a newly allocated tree block, we use implicit back refs for
964 * pointers in it. This means most tree related operations only involve
965 * implicit back refs. For a tree block created in old transaction, the
966 * only way to drop a reference to it is COW it. So we can detect the
967 * event that tree block loses its owner tree's reference and do the
968 * back refs conversion.
970 * When a tree block is COWed through a tree, there are four cases:
972 * The reference count of the block is one and the tree is the block's
973 * owner tree. Nothing to do in this case.
975 * The reference count of the block is one and the tree is not the
976 * block's owner tree. In this case, full back refs is used for pointers
977 * in the block. Remove these full back refs, add implicit back refs for
978 * every pointers in the new block.
980 * The reference count of the block is greater than one and the tree is
981 * the block's owner tree. In this case, implicit back refs is used for
982 * pointers in the block. Add full back refs for every pointers in the
983 * block, increase lower level extents' reference counts. The original
984 * implicit back refs are entailed to the new block.
986 * The reference count of the block is greater than one and the tree is
987 * not the block's owner tree. Add implicit back refs for every pointer in
988 * the new block, increase lower level extents' reference count.
990 * Back Reference Key composing:
992 * The key objectid corresponds to the first byte in the extent,
993 * The key type is used to differentiate between types of back refs.
994 * There are different meanings of the key offset for different types
997 * File extents can be referenced by:
999 * - multiple snapshots, subvolumes, or different generations in one subvol
1000 * - different files inside a single subvolume
1001 * - different offsets inside a file (bookend extents in file.c)
1003 * The extent ref structure for the implicit back refs has fields for:
1005 * - Objectid of the subvolume root
1006 * - objectid of the file holding the reference
1007 * - original offset in the file
1008 * - how many bookend extents
1010 * The key offset for the implicit back refs is hash of the first
1013 * The extent ref structure for the full back refs has field for:
1015 * - number of pointers in the tree leaf
1017 * The key offset for the implicit back refs is the first byte of
1020 * When a file extent is allocated, The implicit back refs is used.
1021 * the fields are filled in:
1023 * (root_key.objectid, inode objectid, offset in file, 1)
1025 * When a file extent is removed file truncation, we find the
1026 * corresponding implicit back refs and check the following fields:
1028 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1030 * Btree extents can be referenced by:
1032 * - Different subvolumes
1034 * Both the implicit back refs and the full back refs for tree blocks
1035 * only consist of key. The key offset for the implicit back refs is
1036 * objectid of block's owner tree. The key offset for the full back refs
1037 * is the first byte of parent block.
1039 * When implicit back refs is used, information about the lowest key and
1040 * level of the tree block are required. These information are stored in
1041 * tree block info structure.
1044 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1045 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1046 struct btrfs_fs_info *fs_info,
1047 struct btrfs_path *path,
1048 u64 owner, u32 extra_size)
1050 struct btrfs_root *root = fs_info->extent_root;
1051 struct btrfs_extent_item *item;
1052 struct btrfs_extent_item_v0 *ei0;
1053 struct btrfs_extent_ref_v0 *ref0;
1054 struct btrfs_tree_block_info *bi;
1055 struct extent_buffer *leaf;
1056 struct btrfs_key key;
1057 struct btrfs_key found_key;
1058 u32 new_size = sizeof(*item);
1062 leaf = path->nodes[0];
1063 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1065 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1066 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1067 struct btrfs_extent_item_v0);
1068 refs = btrfs_extent_refs_v0(leaf, ei0);
1070 if (owner == (u64)-1) {
1072 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1073 ret = btrfs_next_leaf(root, path);
1076 BUG_ON(ret > 0); /* Corruption */
1077 leaf = path->nodes[0];
1079 btrfs_item_key_to_cpu(leaf, &found_key,
1081 BUG_ON(key.objectid != found_key.objectid);
1082 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1086 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1087 struct btrfs_extent_ref_v0);
1088 owner = btrfs_ref_objectid_v0(leaf, ref0);
1092 btrfs_release_path(path);
1094 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1095 new_size += sizeof(*bi);
1097 new_size -= sizeof(*ei0);
1098 ret = btrfs_search_slot(trans, root, &key, path,
1099 new_size + extra_size, 1);
1102 BUG_ON(ret); /* Corruption */
1104 btrfs_extend_item(fs_info, path, new_size);
1106 leaf = path->nodes[0];
1107 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1108 btrfs_set_extent_refs(leaf, item, refs);
1109 /* FIXME: get real generation */
1110 btrfs_set_extent_generation(leaf, item, 0);
1111 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1112 btrfs_set_extent_flags(leaf, item,
1113 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1114 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1115 bi = (struct btrfs_tree_block_info *)(item + 1);
1116 /* FIXME: get first key of the block */
1117 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1118 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1120 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1122 btrfs_mark_buffer_dirty(leaf);
1128 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1129 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1130 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1132 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1133 struct btrfs_extent_inline_ref *iref,
1134 enum btrfs_inline_ref_type is_data)
1136 int type = btrfs_extent_inline_ref_type(eb, iref);
1137 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1139 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1140 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1141 type == BTRFS_SHARED_DATA_REF_KEY ||
1142 type == BTRFS_EXTENT_DATA_REF_KEY) {
1143 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1144 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1146 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1147 ASSERT(eb->fs_info);
1149 * Every shared one has parent tree
1150 * block, which must be aligned to
1154 IS_ALIGNED(offset, eb->fs_info->nodesize))
1157 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1158 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1160 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1161 ASSERT(eb->fs_info);
1163 * Every shared one has parent tree
1164 * block, which must be aligned to
1168 IS_ALIGNED(offset, eb->fs_info->nodesize))
1172 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1177 btrfs_print_leaf((struct extent_buffer *)eb);
1178 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1182 return BTRFS_REF_TYPE_INVALID;
1185 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1187 u32 high_crc = ~(u32)0;
1188 u32 low_crc = ~(u32)0;
1191 lenum = cpu_to_le64(root_objectid);
1192 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1193 lenum = cpu_to_le64(owner);
1194 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1195 lenum = cpu_to_le64(offset);
1196 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1198 return ((u64)high_crc << 31) ^ (u64)low_crc;
1201 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1202 struct btrfs_extent_data_ref *ref)
1204 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1205 btrfs_extent_data_ref_objectid(leaf, ref),
1206 btrfs_extent_data_ref_offset(leaf, ref));
1209 static int match_extent_data_ref(struct extent_buffer *leaf,
1210 struct btrfs_extent_data_ref *ref,
1211 u64 root_objectid, u64 owner, u64 offset)
1213 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1214 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1215 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1220 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1221 struct btrfs_fs_info *fs_info,
1222 struct btrfs_path *path,
1223 u64 bytenr, u64 parent,
1225 u64 owner, u64 offset)
1227 struct btrfs_root *root = fs_info->extent_root;
1228 struct btrfs_key key;
1229 struct btrfs_extent_data_ref *ref;
1230 struct extent_buffer *leaf;
1236 key.objectid = bytenr;
1238 key.type = BTRFS_SHARED_DATA_REF_KEY;
1239 key.offset = parent;
1241 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1242 key.offset = hash_extent_data_ref(root_objectid,
1247 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1256 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1257 key.type = BTRFS_EXTENT_REF_V0_KEY;
1258 btrfs_release_path(path);
1259 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1270 leaf = path->nodes[0];
1271 nritems = btrfs_header_nritems(leaf);
1273 if (path->slots[0] >= nritems) {
1274 ret = btrfs_next_leaf(root, path);
1280 leaf = path->nodes[0];
1281 nritems = btrfs_header_nritems(leaf);
1285 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1286 if (key.objectid != bytenr ||
1287 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1290 ref = btrfs_item_ptr(leaf, path->slots[0],
1291 struct btrfs_extent_data_ref);
1293 if (match_extent_data_ref(leaf, ref, root_objectid,
1296 btrfs_release_path(path);
1308 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1309 struct btrfs_fs_info *fs_info,
1310 struct btrfs_path *path,
1311 u64 bytenr, u64 parent,
1312 u64 root_objectid, u64 owner,
1313 u64 offset, int refs_to_add)
1315 struct btrfs_root *root = fs_info->extent_root;
1316 struct btrfs_key key;
1317 struct extent_buffer *leaf;
1322 key.objectid = bytenr;
1324 key.type = BTRFS_SHARED_DATA_REF_KEY;
1325 key.offset = parent;
1326 size = sizeof(struct btrfs_shared_data_ref);
1328 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1329 key.offset = hash_extent_data_ref(root_objectid,
1331 size = sizeof(struct btrfs_extent_data_ref);
1334 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1335 if (ret && ret != -EEXIST)
1338 leaf = path->nodes[0];
1340 struct btrfs_shared_data_ref *ref;
1341 ref = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_shared_data_ref);
1344 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1347 num_refs += refs_to_add;
1348 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1351 struct btrfs_extent_data_ref *ref;
1352 while (ret == -EEXIST) {
1353 ref = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_extent_data_ref);
1355 if (match_extent_data_ref(leaf, ref, root_objectid,
1358 btrfs_release_path(path);
1360 ret = btrfs_insert_empty_item(trans, root, path, &key,
1362 if (ret && ret != -EEXIST)
1365 leaf = path->nodes[0];
1367 ref = btrfs_item_ptr(leaf, path->slots[0],
1368 struct btrfs_extent_data_ref);
1370 btrfs_set_extent_data_ref_root(leaf, ref,
1372 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1373 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1374 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1376 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1377 num_refs += refs_to_add;
1378 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1381 btrfs_mark_buffer_dirty(leaf);
1384 btrfs_release_path(path);
1388 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1389 struct btrfs_fs_info *fs_info,
1390 struct btrfs_path *path,
1391 int refs_to_drop, int *last_ref)
1393 struct btrfs_key key;
1394 struct btrfs_extent_data_ref *ref1 = NULL;
1395 struct btrfs_shared_data_ref *ref2 = NULL;
1396 struct extent_buffer *leaf;
1400 leaf = path->nodes[0];
1401 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1403 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1404 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1405 struct btrfs_extent_data_ref);
1406 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1407 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1408 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1409 struct btrfs_shared_data_ref);
1410 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1411 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1412 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1413 struct btrfs_extent_ref_v0 *ref0;
1414 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1415 struct btrfs_extent_ref_v0);
1416 num_refs = btrfs_ref_count_v0(leaf, ref0);
1422 BUG_ON(num_refs < refs_to_drop);
1423 num_refs -= refs_to_drop;
1425 if (num_refs == 0) {
1426 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1429 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1430 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1431 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1432 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1433 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1435 struct btrfs_extent_ref_v0 *ref0;
1436 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1437 struct btrfs_extent_ref_v0);
1438 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1441 btrfs_mark_buffer_dirty(leaf);
1446 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1447 struct btrfs_extent_inline_ref *iref)
1449 struct btrfs_key key;
1450 struct extent_buffer *leaf;
1451 struct btrfs_extent_data_ref *ref1;
1452 struct btrfs_shared_data_ref *ref2;
1456 leaf = path->nodes[0];
1457 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1460 * If type is invalid, we should have bailed out earlier than
1463 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1464 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1465 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1466 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1467 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1469 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1470 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1472 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1473 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1474 struct btrfs_extent_data_ref);
1475 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1476 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1477 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1478 struct btrfs_shared_data_ref);
1479 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1480 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1481 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1482 struct btrfs_extent_ref_v0 *ref0;
1483 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1484 struct btrfs_extent_ref_v0);
1485 num_refs = btrfs_ref_count_v0(leaf, ref0);
1493 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1494 struct btrfs_fs_info *fs_info,
1495 struct btrfs_path *path,
1496 u64 bytenr, u64 parent,
1499 struct btrfs_root *root = fs_info->extent_root;
1500 struct btrfs_key key;
1503 key.objectid = bytenr;
1505 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1506 key.offset = parent;
1508 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1509 key.offset = root_objectid;
1512 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1515 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1516 if (ret == -ENOENT && parent) {
1517 btrfs_release_path(path);
1518 key.type = BTRFS_EXTENT_REF_V0_KEY;
1519 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1527 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1528 struct btrfs_fs_info *fs_info,
1529 struct btrfs_path *path,
1530 u64 bytenr, u64 parent,
1533 struct btrfs_key key;
1536 key.objectid = bytenr;
1538 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1539 key.offset = parent;
1541 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1542 key.offset = root_objectid;
1545 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1547 btrfs_release_path(path);
1551 static inline int extent_ref_type(u64 parent, u64 owner)
1554 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1556 type = BTRFS_SHARED_BLOCK_REF_KEY;
1558 type = BTRFS_TREE_BLOCK_REF_KEY;
1561 type = BTRFS_SHARED_DATA_REF_KEY;
1563 type = BTRFS_EXTENT_DATA_REF_KEY;
1568 static int find_next_key(struct btrfs_path *path, int level,
1569 struct btrfs_key *key)
1572 for (; level < BTRFS_MAX_LEVEL; level++) {
1573 if (!path->nodes[level])
1575 if (path->slots[level] + 1 >=
1576 btrfs_header_nritems(path->nodes[level]))
1579 btrfs_item_key_to_cpu(path->nodes[level], key,
1580 path->slots[level] + 1);
1582 btrfs_node_key_to_cpu(path->nodes[level], key,
1583 path->slots[level] + 1);
1590 * look for inline back ref. if back ref is found, *ref_ret is set
1591 * to the address of inline back ref, and 0 is returned.
1593 * if back ref isn't found, *ref_ret is set to the address where it
1594 * should be inserted, and -ENOENT is returned.
1596 * if insert is true and there are too many inline back refs, the path
1597 * points to the extent item, and -EAGAIN is returned.
1599 * NOTE: inline back refs are ordered in the same way that back ref
1600 * items in the tree are ordered.
1602 static noinline_for_stack
1603 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1604 struct btrfs_fs_info *fs_info,
1605 struct btrfs_path *path,
1606 struct btrfs_extent_inline_ref **ref_ret,
1607 u64 bytenr, u64 num_bytes,
1608 u64 parent, u64 root_objectid,
1609 u64 owner, u64 offset, int insert)
1611 struct btrfs_root *root = fs_info->extent_root;
1612 struct btrfs_key key;
1613 struct extent_buffer *leaf;
1614 struct btrfs_extent_item *ei;
1615 struct btrfs_extent_inline_ref *iref;
1625 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1628 key.objectid = bytenr;
1629 key.type = BTRFS_EXTENT_ITEM_KEY;
1630 key.offset = num_bytes;
1632 want = extent_ref_type(parent, owner);
1634 extra_size = btrfs_extent_inline_ref_size(want);
1635 path->keep_locks = 1;
1640 * Owner is our parent level, so we can just add one to get the level
1641 * for the block we are interested in.
1643 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1644 key.type = BTRFS_METADATA_ITEM_KEY;
1649 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1656 * We may be a newly converted file system which still has the old fat
1657 * extent entries for metadata, so try and see if we have one of those.
1659 if (ret > 0 && skinny_metadata) {
1660 skinny_metadata = false;
1661 if (path->slots[0]) {
1663 btrfs_item_key_to_cpu(path->nodes[0], &key,
1665 if (key.objectid == bytenr &&
1666 key.type == BTRFS_EXTENT_ITEM_KEY &&
1667 key.offset == num_bytes)
1671 key.objectid = bytenr;
1672 key.type = BTRFS_EXTENT_ITEM_KEY;
1673 key.offset = num_bytes;
1674 btrfs_release_path(path);
1679 if (ret && !insert) {
1682 } else if (WARN_ON(ret)) {
1687 leaf = path->nodes[0];
1688 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1689 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1690 if (item_size < sizeof(*ei)) {
1695 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1701 leaf = path->nodes[0];
1702 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1705 BUG_ON(item_size < sizeof(*ei));
1707 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1708 flags = btrfs_extent_flags(leaf, ei);
1710 ptr = (unsigned long)(ei + 1);
1711 end = (unsigned long)ei + item_size;
1713 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1714 ptr += sizeof(struct btrfs_tree_block_info);
1718 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1719 needed = BTRFS_REF_TYPE_DATA;
1721 needed = BTRFS_REF_TYPE_BLOCK;
1729 iref = (struct btrfs_extent_inline_ref *)ptr;
1730 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1731 if (type == BTRFS_REF_TYPE_INVALID) {
1739 ptr += btrfs_extent_inline_ref_size(type);
1743 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1744 struct btrfs_extent_data_ref *dref;
1745 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1746 if (match_extent_data_ref(leaf, dref, root_objectid,
1751 if (hash_extent_data_ref_item(leaf, dref) <
1752 hash_extent_data_ref(root_objectid, owner, offset))
1756 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1758 if (parent == ref_offset) {
1762 if (ref_offset < parent)
1765 if (root_objectid == ref_offset) {
1769 if (ref_offset < root_objectid)
1773 ptr += btrfs_extent_inline_ref_size(type);
1775 if (err == -ENOENT && insert) {
1776 if (item_size + extra_size >=
1777 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1782 * To add new inline back ref, we have to make sure
1783 * there is no corresponding back ref item.
1784 * For simplicity, we just do not add new inline back
1785 * ref if there is any kind of item for this block
1787 if (find_next_key(path, 0, &key) == 0 &&
1788 key.objectid == bytenr &&
1789 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1794 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1797 path->keep_locks = 0;
1798 btrfs_unlock_up_safe(path, 1);
1804 * helper to add new inline back ref
1806 static noinline_for_stack
1807 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1808 struct btrfs_path *path,
1809 struct btrfs_extent_inline_ref *iref,
1810 u64 parent, u64 root_objectid,
1811 u64 owner, u64 offset, int refs_to_add,
1812 struct btrfs_delayed_extent_op *extent_op)
1814 struct extent_buffer *leaf;
1815 struct btrfs_extent_item *ei;
1818 unsigned long item_offset;
1823 leaf = path->nodes[0];
1824 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1825 item_offset = (unsigned long)iref - (unsigned long)ei;
1827 type = extent_ref_type(parent, owner);
1828 size = btrfs_extent_inline_ref_size(type);
1830 btrfs_extend_item(fs_info, path, size);
1832 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1833 refs = btrfs_extent_refs(leaf, ei);
1834 refs += refs_to_add;
1835 btrfs_set_extent_refs(leaf, ei, refs);
1837 __run_delayed_extent_op(extent_op, leaf, ei);
1839 ptr = (unsigned long)ei + item_offset;
1840 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1841 if (ptr < end - size)
1842 memmove_extent_buffer(leaf, ptr + size, ptr,
1845 iref = (struct btrfs_extent_inline_ref *)ptr;
1846 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1847 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1848 struct btrfs_extent_data_ref *dref;
1849 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1850 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1851 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1852 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1853 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1854 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1855 struct btrfs_shared_data_ref *sref;
1856 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1857 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1858 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1859 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1860 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1862 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1864 btrfs_mark_buffer_dirty(leaf);
1867 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1868 struct btrfs_fs_info *fs_info,
1869 struct btrfs_path *path,
1870 struct btrfs_extent_inline_ref **ref_ret,
1871 u64 bytenr, u64 num_bytes, u64 parent,
1872 u64 root_objectid, u64 owner, u64 offset)
1876 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1877 bytenr, num_bytes, parent,
1878 root_objectid, owner, offset, 0);
1882 btrfs_release_path(path);
1885 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1886 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1887 parent, root_objectid);
1889 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1890 parent, root_objectid, owner,
1897 * helper to update/remove inline back ref
1899 static noinline_for_stack
1900 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1901 struct btrfs_path *path,
1902 struct btrfs_extent_inline_ref *iref,
1904 struct btrfs_delayed_extent_op *extent_op,
1907 struct extent_buffer *leaf;
1908 struct btrfs_extent_item *ei;
1909 struct btrfs_extent_data_ref *dref = NULL;
1910 struct btrfs_shared_data_ref *sref = NULL;
1918 leaf = path->nodes[0];
1919 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1920 refs = btrfs_extent_refs(leaf, ei);
1921 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1922 refs += refs_to_mod;
1923 btrfs_set_extent_refs(leaf, ei, refs);
1925 __run_delayed_extent_op(extent_op, leaf, ei);
1928 * If type is invalid, we should have bailed out after
1929 * lookup_inline_extent_backref().
1931 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1932 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1934 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1935 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1936 refs = btrfs_extent_data_ref_count(leaf, dref);
1937 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1938 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1939 refs = btrfs_shared_data_ref_count(leaf, sref);
1942 BUG_ON(refs_to_mod != -1);
1945 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1946 refs += refs_to_mod;
1949 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1950 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1952 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1955 size = btrfs_extent_inline_ref_size(type);
1956 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1957 ptr = (unsigned long)iref;
1958 end = (unsigned long)ei + item_size;
1959 if (ptr + size < end)
1960 memmove_extent_buffer(leaf, ptr, ptr + size,
1963 btrfs_truncate_item(fs_info, path, item_size, 1);
1965 btrfs_mark_buffer_dirty(leaf);
1968 static noinline_for_stack
1969 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1970 struct btrfs_fs_info *fs_info,
1971 struct btrfs_path *path,
1972 u64 bytenr, u64 num_bytes, u64 parent,
1973 u64 root_objectid, u64 owner,
1974 u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_extent_inline_ref *iref;
1980 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1981 bytenr, num_bytes, parent,
1982 root_objectid, owner, offset, 1);
1984 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1985 update_inline_extent_backref(fs_info, path, iref,
1986 refs_to_add, extent_op, NULL);
1987 } else if (ret == -ENOENT) {
1988 setup_inline_extent_backref(fs_info, path, iref, parent,
1989 root_objectid, owner, offset,
1990 refs_to_add, extent_op);
1996 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1997 struct btrfs_fs_info *fs_info,
1998 struct btrfs_path *path,
1999 u64 bytenr, u64 parent, u64 root_objectid,
2000 u64 owner, u64 offset, int refs_to_add)
2003 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2004 BUG_ON(refs_to_add != 1);
2005 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2006 parent, root_objectid);
2008 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2009 parent, root_objectid,
2010 owner, offset, refs_to_add);
2015 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2016 struct btrfs_fs_info *fs_info,
2017 struct btrfs_path *path,
2018 struct btrfs_extent_inline_ref *iref,
2019 int refs_to_drop, int is_data, int *last_ref)
2023 BUG_ON(!is_data && refs_to_drop != 1);
2025 update_inline_extent_backref(fs_info, path, iref,
2026 -refs_to_drop, NULL, last_ref);
2027 } else if (is_data) {
2028 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2032 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2037 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2038 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2039 u64 *discarded_bytes)
2042 u64 bytes_left, end;
2043 u64 aligned_start = ALIGN(start, 1 << 9);
2045 if (WARN_ON(start != aligned_start)) {
2046 len -= aligned_start - start;
2047 len = round_down(len, 1 << 9);
2048 start = aligned_start;
2051 *discarded_bytes = 0;
2059 /* Skip any superblocks on this device. */
2060 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2061 u64 sb_start = btrfs_sb_offset(j);
2062 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2063 u64 size = sb_start - start;
2065 if (!in_range(sb_start, start, bytes_left) &&
2066 !in_range(sb_end, start, bytes_left) &&
2067 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2071 * Superblock spans beginning of range. Adjust start and
2074 if (sb_start <= start) {
2075 start += sb_end - start;
2080 bytes_left = end - start;
2085 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2088 *discarded_bytes += size;
2089 else if (ret != -EOPNOTSUPP)
2098 bytes_left = end - start;
2102 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2105 *discarded_bytes += bytes_left;
2110 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2111 u64 num_bytes, u64 *actual_bytes)
2114 u64 discarded_bytes = 0;
2115 struct btrfs_bio *bbio = NULL;
2119 * Avoid races with device replace and make sure our bbio has devices
2120 * associated to its stripes that don't go away while we are discarding.
2122 btrfs_bio_counter_inc_blocked(fs_info);
2123 /* Tell the block device(s) that the sectors can be discarded */
2124 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2126 /* Error condition is -ENOMEM */
2128 struct btrfs_bio_stripe *stripe = bbio->stripes;
2132 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2134 struct request_queue *req_q;
2136 if (!stripe->dev->bdev) {
2137 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2140 req_q = bdev_get_queue(stripe->dev->bdev);
2141 if (!blk_queue_discard(req_q))
2144 ret = btrfs_issue_discard(stripe->dev->bdev,
2149 discarded_bytes += bytes;
2150 else if (ret != -EOPNOTSUPP)
2151 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2154 * Just in case we get back EOPNOTSUPP for some reason,
2155 * just ignore the return value so we don't screw up
2156 * people calling discard_extent.
2160 btrfs_put_bbio(bbio);
2162 btrfs_bio_counter_dec(fs_info);
2165 *actual_bytes = discarded_bytes;
2168 if (ret == -EOPNOTSUPP)
2173 /* Can return -ENOMEM */
2174 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2175 struct btrfs_root *root,
2176 u64 bytenr, u64 num_bytes, u64 parent,
2177 u64 root_objectid, u64 owner, u64 offset)
2179 struct btrfs_fs_info *fs_info = root->fs_info;
2180 int old_ref_mod, new_ref_mod;
2183 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2184 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2186 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2187 owner, offset, BTRFS_ADD_DELAYED_REF);
2189 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2190 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2192 root_objectid, (int)owner,
2193 BTRFS_ADD_DELAYED_REF, NULL,
2194 &old_ref_mod, &new_ref_mod);
2196 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2198 root_objectid, owner, offset,
2199 0, BTRFS_ADD_DELAYED_REF,
2200 &old_ref_mod, &new_ref_mod);
2203 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2204 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2209 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2210 struct btrfs_fs_info *fs_info,
2211 struct btrfs_delayed_ref_node *node,
2212 u64 parent, u64 root_objectid,
2213 u64 owner, u64 offset, int refs_to_add,
2214 struct btrfs_delayed_extent_op *extent_op)
2216 struct btrfs_path *path;
2217 struct extent_buffer *leaf;
2218 struct btrfs_extent_item *item;
2219 struct btrfs_key key;
2220 u64 bytenr = node->bytenr;
2221 u64 num_bytes = node->num_bytes;
2225 path = btrfs_alloc_path();
2229 path->reada = READA_FORWARD;
2230 path->leave_spinning = 1;
2231 /* this will setup the path even if it fails to insert the back ref */
2232 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2233 num_bytes, parent, root_objectid,
2235 refs_to_add, extent_op);
2236 if ((ret < 0 && ret != -EAGAIN) || !ret)
2240 * Ok we had -EAGAIN which means we didn't have space to insert and
2241 * inline extent ref, so just update the reference count and add a
2244 leaf = path->nodes[0];
2245 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2246 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2247 refs = btrfs_extent_refs(leaf, item);
2248 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2250 __run_delayed_extent_op(extent_op, leaf, item);
2252 btrfs_mark_buffer_dirty(leaf);
2253 btrfs_release_path(path);
2255 path->reada = READA_FORWARD;
2256 path->leave_spinning = 1;
2257 /* now insert the actual backref */
2258 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2259 root_objectid, owner, offset, refs_to_add);
2261 btrfs_abort_transaction(trans, ret);
2263 btrfs_free_path(path);
2267 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2268 struct btrfs_fs_info *fs_info,
2269 struct btrfs_delayed_ref_node *node,
2270 struct btrfs_delayed_extent_op *extent_op,
2271 int insert_reserved)
2274 struct btrfs_delayed_data_ref *ref;
2275 struct btrfs_key ins;
2280 ins.objectid = node->bytenr;
2281 ins.offset = node->num_bytes;
2282 ins.type = BTRFS_EXTENT_ITEM_KEY;
2284 ref = btrfs_delayed_node_to_data_ref(node);
2285 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2287 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2288 parent = ref->parent;
2289 ref_root = ref->root;
2291 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2293 flags |= extent_op->flags_to_set;
2294 ret = alloc_reserved_file_extent(trans, fs_info,
2295 parent, ref_root, flags,
2296 ref->objectid, ref->offset,
2297 &ins, node->ref_mod);
2298 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2299 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2300 ref_root, ref->objectid,
2301 ref->offset, node->ref_mod,
2303 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2304 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2305 ref_root, ref->objectid,
2306 ref->offset, node->ref_mod,
2314 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2315 struct extent_buffer *leaf,
2316 struct btrfs_extent_item *ei)
2318 u64 flags = btrfs_extent_flags(leaf, ei);
2319 if (extent_op->update_flags) {
2320 flags |= extent_op->flags_to_set;
2321 btrfs_set_extent_flags(leaf, ei, flags);
2324 if (extent_op->update_key) {
2325 struct btrfs_tree_block_info *bi;
2326 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2327 bi = (struct btrfs_tree_block_info *)(ei + 1);
2328 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2332 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2333 struct btrfs_fs_info *fs_info,
2334 struct btrfs_delayed_ref_head *head,
2335 struct btrfs_delayed_extent_op *extent_op)
2337 struct btrfs_key key;
2338 struct btrfs_path *path;
2339 struct btrfs_extent_item *ei;
2340 struct extent_buffer *leaf;
2344 int metadata = !extent_op->is_data;
2349 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2352 path = btrfs_alloc_path();
2356 key.objectid = head->bytenr;
2359 key.type = BTRFS_METADATA_ITEM_KEY;
2360 key.offset = extent_op->level;
2362 key.type = BTRFS_EXTENT_ITEM_KEY;
2363 key.offset = head->num_bytes;
2367 path->reada = READA_FORWARD;
2368 path->leave_spinning = 1;
2369 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2376 if (path->slots[0] > 0) {
2378 btrfs_item_key_to_cpu(path->nodes[0], &key,
2380 if (key.objectid == head->bytenr &&
2381 key.type == BTRFS_EXTENT_ITEM_KEY &&
2382 key.offset == head->num_bytes)
2386 btrfs_release_path(path);
2389 key.objectid = head->bytenr;
2390 key.offset = head->num_bytes;
2391 key.type = BTRFS_EXTENT_ITEM_KEY;
2400 leaf = path->nodes[0];
2401 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2402 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2403 if (item_size < sizeof(*ei)) {
2404 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2409 leaf = path->nodes[0];
2410 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2413 BUG_ON(item_size < sizeof(*ei));
2414 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2415 __run_delayed_extent_op(extent_op, leaf, ei);
2417 btrfs_mark_buffer_dirty(leaf);
2419 btrfs_free_path(path);
2423 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2424 struct btrfs_fs_info *fs_info,
2425 struct btrfs_delayed_ref_node *node,
2426 struct btrfs_delayed_extent_op *extent_op,
2427 int insert_reserved)
2430 struct btrfs_delayed_tree_ref *ref;
2431 struct btrfs_key ins;
2434 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2436 ref = btrfs_delayed_node_to_tree_ref(node);
2437 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2439 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2440 parent = ref->parent;
2441 ref_root = ref->root;
2443 ins.objectid = node->bytenr;
2444 if (skinny_metadata) {
2445 ins.offset = ref->level;
2446 ins.type = BTRFS_METADATA_ITEM_KEY;
2448 ins.offset = node->num_bytes;
2449 ins.type = BTRFS_EXTENT_ITEM_KEY;
2452 if (node->ref_mod != 1) {
2454 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2455 node->bytenr, node->ref_mod, node->action, ref_root,
2459 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2460 BUG_ON(!extent_op || !extent_op->update_flags);
2461 ret = alloc_reserved_tree_block(trans, fs_info,
2463 extent_op->flags_to_set,
2466 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2467 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2471 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2472 ret = __btrfs_free_extent(trans, fs_info, node,
2474 ref->level, 0, 1, extent_op);
2481 /* helper function to actually process a single delayed ref entry */
2482 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2483 struct btrfs_fs_info *fs_info,
2484 struct btrfs_delayed_ref_node *node,
2485 struct btrfs_delayed_extent_op *extent_op,
2486 int insert_reserved)
2490 if (trans->aborted) {
2491 if (insert_reserved)
2492 btrfs_pin_extent(fs_info, node->bytenr,
2493 node->num_bytes, 1);
2497 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2498 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2499 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2501 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2502 node->type == BTRFS_SHARED_DATA_REF_KEY)
2503 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2510 static inline struct btrfs_delayed_ref_node *
2511 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2513 struct btrfs_delayed_ref_node *ref;
2515 if (RB_EMPTY_ROOT(&head->ref_tree))
2519 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2520 * This is to prevent a ref count from going down to zero, which deletes
2521 * the extent item from the extent tree, when there still are references
2522 * to add, which would fail because they would not find the extent item.
2524 if (!list_empty(&head->ref_add_list))
2525 return list_first_entry(&head->ref_add_list,
2526 struct btrfs_delayed_ref_node, add_list);
2528 ref = rb_entry(rb_first(&head->ref_tree),
2529 struct btrfs_delayed_ref_node, ref_node);
2530 ASSERT(list_empty(&ref->add_list));
2534 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2535 struct btrfs_delayed_ref_head *head)
2537 spin_lock(&delayed_refs->lock);
2538 head->processing = 0;
2539 delayed_refs->num_heads_ready++;
2540 spin_unlock(&delayed_refs->lock);
2541 btrfs_delayed_ref_unlock(head);
2544 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2545 struct btrfs_fs_info *fs_info,
2546 struct btrfs_delayed_ref_head *head)
2548 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2553 head->extent_op = NULL;
2554 if (head->must_insert_reserved) {
2555 btrfs_free_delayed_extent_op(extent_op);
2558 spin_unlock(&head->lock);
2559 ret = run_delayed_extent_op(trans, fs_info, head, extent_op);
2560 btrfs_free_delayed_extent_op(extent_op);
2561 return ret ? ret : 1;
2564 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2565 struct btrfs_fs_info *fs_info,
2566 struct btrfs_delayed_ref_head *head)
2568 struct btrfs_delayed_ref_root *delayed_refs;
2571 delayed_refs = &trans->transaction->delayed_refs;
2573 ret = cleanup_extent_op(trans, fs_info, head);
2575 unselect_delayed_ref_head(delayed_refs, head);
2576 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2583 * Need to drop our head ref lock and re-acquire the delayed ref lock
2584 * and then re-check to make sure nobody got added.
2586 spin_unlock(&head->lock);
2587 spin_lock(&delayed_refs->lock);
2588 spin_lock(&head->lock);
2589 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2590 spin_unlock(&head->lock);
2591 spin_unlock(&delayed_refs->lock);
2594 delayed_refs->num_heads--;
2595 rb_erase(&head->href_node, &delayed_refs->href_root);
2596 RB_CLEAR_NODE(&head->href_node);
2597 spin_unlock(&delayed_refs->lock);
2598 spin_unlock(&head->lock);
2599 atomic_dec(&delayed_refs->num_entries);
2601 trace_run_delayed_ref_head(fs_info, head, 0);
2603 if (head->total_ref_mod < 0) {
2604 struct btrfs_block_group_cache *cache;
2606 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
2608 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2610 btrfs_put_block_group(cache);
2612 if (head->is_data) {
2613 spin_lock(&delayed_refs->lock);
2614 delayed_refs->pending_csums -= head->num_bytes;
2615 spin_unlock(&delayed_refs->lock);
2619 if (head->must_insert_reserved) {
2620 btrfs_pin_extent(fs_info, head->bytenr,
2621 head->num_bytes, 1);
2622 if (head->is_data) {
2623 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2628 /* Also free its reserved qgroup space */
2629 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2630 head->qgroup_reserved);
2631 btrfs_delayed_ref_unlock(head);
2632 btrfs_put_delayed_ref_head(head);
2637 * Returns 0 on success or if called with an already aborted transaction.
2638 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2640 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2643 struct btrfs_fs_info *fs_info = trans->fs_info;
2644 struct btrfs_delayed_ref_root *delayed_refs;
2645 struct btrfs_delayed_ref_node *ref;
2646 struct btrfs_delayed_ref_head *locked_ref = NULL;
2647 struct btrfs_delayed_extent_op *extent_op;
2648 ktime_t start = ktime_get();
2650 unsigned long count = 0;
2651 unsigned long actual_count = 0;
2652 int must_insert_reserved = 0;
2654 delayed_refs = &trans->transaction->delayed_refs;
2660 spin_lock(&delayed_refs->lock);
2661 locked_ref = btrfs_select_ref_head(trans);
2663 spin_unlock(&delayed_refs->lock);
2667 /* grab the lock that says we are going to process
2668 * all the refs for this head */
2669 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2670 spin_unlock(&delayed_refs->lock);
2672 * we may have dropped the spin lock to get the head
2673 * mutex lock, and that might have given someone else
2674 * time to free the head. If that's true, it has been
2675 * removed from our list and we can move on.
2677 if (ret == -EAGAIN) {
2685 * We need to try and merge add/drops of the same ref since we
2686 * can run into issues with relocate dropping the implicit ref
2687 * and then it being added back again before the drop can
2688 * finish. If we merged anything we need to re-loop so we can
2690 * Or we can get node references of the same type that weren't
2691 * merged when created due to bumps in the tree mod seq, and
2692 * we need to merge them to prevent adding an inline extent
2693 * backref before dropping it (triggering a BUG_ON at
2694 * insert_inline_extent_backref()).
2696 spin_lock(&locked_ref->lock);
2697 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2701 * locked_ref is the head node, so we have to go one
2702 * node back for any delayed ref updates
2704 ref = select_delayed_ref(locked_ref);
2706 if (ref && ref->seq &&
2707 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2708 spin_unlock(&locked_ref->lock);
2709 unselect_delayed_ref_head(delayed_refs, locked_ref);
2717 * We're done processing refs in this ref_head, clean everything
2718 * up and move on to the next ref_head.
2721 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2723 /* We dropped our lock, we need to loop. */
2736 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2737 RB_CLEAR_NODE(&ref->ref_node);
2738 if (!list_empty(&ref->add_list))
2739 list_del(&ref->add_list);
2741 * When we play the delayed ref, also correct the ref_mod on
2744 switch (ref->action) {
2745 case BTRFS_ADD_DELAYED_REF:
2746 case BTRFS_ADD_DELAYED_EXTENT:
2747 locked_ref->ref_mod -= ref->ref_mod;
2749 case BTRFS_DROP_DELAYED_REF:
2750 locked_ref->ref_mod += ref->ref_mod;
2755 atomic_dec(&delayed_refs->num_entries);
2758 * Record the must-insert_reserved flag before we drop the spin
2761 must_insert_reserved = locked_ref->must_insert_reserved;
2762 locked_ref->must_insert_reserved = 0;
2764 extent_op = locked_ref->extent_op;
2765 locked_ref->extent_op = NULL;
2766 spin_unlock(&locked_ref->lock);
2768 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2769 must_insert_reserved);
2771 btrfs_free_delayed_extent_op(extent_op);
2773 unselect_delayed_ref_head(delayed_refs, locked_ref);
2774 btrfs_put_delayed_ref(ref);
2775 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2780 btrfs_put_delayed_ref(ref);
2786 * We don't want to include ref heads since we can have empty ref heads
2787 * and those will drastically skew our runtime down since we just do
2788 * accounting, no actual extent tree updates.
2790 if (actual_count > 0) {
2791 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2795 * We weigh the current average higher than our current runtime
2796 * to avoid large swings in the average.
2798 spin_lock(&delayed_refs->lock);
2799 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2800 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2801 spin_unlock(&delayed_refs->lock);
2806 #ifdef SCRAMBLE_DELAYED_REFS
2808 * Normally delayed refs get processed in ascending bytenr order. This
2809 * correlates in most cases to the order added. To expose dependencies on this
2810 * order, we start to process the tree in the middle instead of the beginning
2812 static u64 find_middle(struct rb_root *root)
2814 struct rb_node *n = root->rb_node;
2815 struct btrfs_delayed_ref_node *entry;
2818 u64 first = 0, last = 0;
2822 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2823 first = entry->bytenr;
2827 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2828 last = entry->bytenr;
2833 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2834 WARN_ON(!entry->in_tree);
2836 middle = entry->bytenr;
2849 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2853 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2854 sizeof(struct btrfs_extent_inline_ref));
2855 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2856 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2859 * We don't ever fill up leaves all the way so multiply by 2 just to be
2860 * closer to what we're really going to want to use.
2862 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2866 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2867 * would require to store the csums for that many bytes.
2869 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2872 u64 num_csums_per_leaf;
2875 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2876 num_csums_per_leaf = div64_u64(csum_size,
2877 (u64)btrfs_super_csum_size(fs_info->super_copy));
2878 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2879 num_csums += num_csums_per_leaf - 1;
2880 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2884 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2885 struct btrfs_fs_info *fs_info)
2887 struct btrfs_block_rsv *global_rsv;
2888 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2889 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2890 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2891 u64 num_bytes, num_dirty_bgs_bytes;
2894 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2895 num_heads = heads_to_leaves(fs_info, num_heads);
2897 num_bytes += (num_heads - 1) * fs_info->nodesize;
2899 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2901 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2903 global_rsv = &fs_info->global_block_rsv;
2906 * If we can't allocate any more chunks lets make sure we have _lots_ of
2907 * wiggle room since running delayed refs can create more delayed refs.
2909 if (global_rsv->space_info->full) {
2910 num_dirty_bgs_bytes <<= 1;
2914 spin_lock(&global_rsv->lock);
2915 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2917 spin_unlock(&global_rsv->lock);
2921 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2922 struct btrfs_fs_info *fs_info)
2925 atomic_read(&trans->transaction->delayed_refs.num_entries);
2930 avg_runtime = fs_info->avg_delayed_ref_runtime;
2931 val = num_entries * avg_runtime;
2932 if (val >= NSEC_PER_SEC)
2934 if (val >= NSEC_PER_SEC / 2)
2937 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2940 struct async_delayed_refs {
2941 struct btrfs_root *root;
2946 struct completion wait;
2947 struct btrfs_work work;
2950 static inline struct async_delayed_refs *
2951 to_async_delayed_refs(struct btrfs_work *work)
2953 return container_of(work, struct async_delayed_refs, work);
2956 static void delayed_ref_async_start(struct btrfs_work *work)
2958 struct async_delayed_refs *async = to_async_delayed_refs(work);
2959 struct btrfs_trans_handle *trans;
2960 struct btrfs_fs_info *fs_info = async->root->fs_info;
2963 /* if the commit is already started, we don't need to wait here */
2964 if (btrfs_transaction_blocked(fs_info))
2967 trans = btrfs_join_transaction(async->root);
2968 if (IS_ERR(trans)) {
2969 async->error = PTR_ERR(trans);
2974 * trans->sync means that when we call end_transaction, we won't
2975 * wait on delayed refs
2979 /* Don't bother flushing if we got into a different transaction */
2980 if (trans->transid > async->transid)
2983 ret = btrfs_run_delayed_refs(trans, async->count);
2987 ret = btrfs_end_transaction(trans);
2988 if (ret && !async->error)
2992 complete(&async->wait);
2997 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2998 unsigned long count, u64 transid, int wait)
3000 struct async_delayed_refs *async;
3003 async = kmalloc(sizeof(*async), GFP_NOFS);
3007 async->root = fs_info->tree_root;
3008 async->count = count;
3010 async->transid = transid;
3015 init_completion(&async->wait);
3017 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3018 delayed_ref_async_start, NULL, NULL);
3020 btrfs_queue_work(fs_info->extent_workers, &async->work);
3023 wait_for_completion(&async->wait);
3032 * this starts processing the delayed reference count updates and
3033 * extent insertions we have queued up so far. count can be
3034 * 0, which means to process everything in the tree at the start
3035 * of the run (but not newly added entries), or it can be some target
3036 * number you'd like to process.
3038 * Returns 0 on success or if called with an aborted transaction
3039 * Returns <0 on error and aborts the transaction
3041 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3042 unsigned long count)
3044 struct btrfs_fs_info *fs_info = trans->fs_info;
3045 struct rb_node *node;
3046 struct btrfs_delayed_ref_root *delayed_refs;
3047 struct btrfs_delayed_ref_head *head;
3049 int run_all = count == (unsigned long)-1;
3050 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3052 /* We'll clean this up in btrfs_cleanup_transaction */
3056 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3059 delayed_refs = &trans->transaction->delayed_refs;
3061 count = atomic_read(&delayed_refs->num_entries) * 2;
3064 #ifdef SCRAMBLE_DELAYED_REFS
3065 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3067 trans->can_flush_pending_bgs = false;
3068 ret = __btrfs_run_delayed_refs(trans, count);
3070 btrfs_abort_transaction(trans, ret);
3075 if (!list_empty(&trans->new_bgs))
3076 btrfs_create_pending_block_groups(trans);
3078 spin_lock(&delayed_refs->lock);
3079 node = rb_first(&delayed_refs->href_root);
3081 spin_unlock(&delayed_refs->lock);
3084 head = rb_entry(node, struct btrfs_delayed_ref_head,
3086 refcount_inc(&head->refs);
3087 spin_unlock(&delayed_refs->lock);
3089 /* Mutex was contended, block until it's released and retry. */
3090 mutex_lock(&head->mutex);
3091 mutex_unlock(&head->mutex);
3093 btrfs_put_delayed_ref_head(head);
3098 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3102 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3103 struct btrfs_fs_info *fs_info,
3104 u64 bytenr, u64 num_bytes, u64 flags,
3105 int level, int is_data)
3107 struct btrfs_delayed_extent_op *extent_op;
3110 extent_op = btrfs_alloc_delayed_extent_op();
3114 extent_op->flags_to_set = flags;
3115 extent_op->update_flags = true;
3116 extent_op->update_key = false;
3117 extent_op->is_data = is_data ? true : false;
3118 extent_op->level = level;
3120 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3121 num_bytes, extent_op);
3123 btrfs_free_delayed_extent_op(extent_op);
3127 static noinline int check_delayed_ref(struct btrfs_root *root,
3128 struct btrfs_path *path,
3129 u64 objectid, u64 offset, u64 bytenr)
3131 struct btrfs_delayed_ref_head *head;
3132 struct btrfs_delayed_ref_node *ref;
3133 struct btrfs_delayed_data_ref *data_ref;
3134 struct btrfs_delayed_ref_root *delayed_refs;
3135 struct btrfs_transaction *cur_trans;
3136 struct rb_node *node;
3139 cur_trans = root->fs_info->running_transaction;
3143 delayed_refs = &cur_trans->delayed_refs;
3144 spin_lock(&delayed_refs->lock);
3145 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3147 spin_unlock(&delayed_refs->lock);
3151 if (!mutex_trylock(&head->mutex)) {
3152 refcount_inc(&head->refs);
3153 spin_unlock(&delayed_refs->lock);
3155 btrfs_release_path(path);
3158 * Mutex was contended, block until it's released and let
3161 mutex_lock(&head->mutex);
3162 mutex_unlock(&head->mutex);
3163 btrfs_put_delayed_ref_head(head);
3166 spin_unlock(&delayed_refs->lock);
3168 spin_lock(&head->lock);
3170 * XXX: We should replace this with a proper search function in the
3173 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3174 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3175 /* If it's a shared ref we know a cross reference exists */
3176 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3181 data_ref = btrfs_delayed_node_to_data_ref(ref);
3184 * If our ref doesn't match the one we're currently looking at
3185 * then we have a cross reference.
3187 if (data_ref->root != root->root_key.objectid ||
3188 data_ref->objectid != objectid ||
3189 data_ref->offset != offset) {
3194 spin_unlock(&head->lock);
3195 mutex_unlock(&head->mutex);
3199 static noinline int check_committed_ref(struct btrfs_root *root,
3200 struct btrfs_path *path,
3201 u64 objectid, u64 offset, u64 bytenr)
3203 struct btrfs_fs_info *fs_info = root->fs_info;
3204 struct btrfs_root *extent_root = fs_info->extent_root;
3205 struct extent_buffer *leaf;
3206 struct btrfs_extent_data_ref *ref;
3207 struct btrfs_extent_inline_ref *iref;
3208 struct btrfs_extent_item *ei;
3209 struct btrfs_key key;
3214 key.objectid = bytenr;
3215 key.offset = (u64)-1;
3216 key.type = BTRFS_EXTENT_ITEM_KEY;
3218 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3221 BUG_ON(ret == 0); /* Corruption */
3224 if (path->slots[0] == 0)
3228 leaf = path->nodes[0];
3229 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3231 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3235 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3236 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3237 if (item_size < sizeof(*ei)) {
3238 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3242 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3244 if (item_size != sizeof(*ei) +
3245 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3248 if (btrfs_extent_generation(leaf, ei) <=
3249 btrfs_root_last_snapshot(&root->root_item))
3252 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3254 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3255 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3258 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3259 if (btrfs_extent_refs(leaf, ei) !=
3260 btrfs_extent_data_ref_count(leaf, ref) ||
3261 btrfs_extent_data_ref_root(leaf, ref) !=
3262 root->root_key.objectid ||
3263 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3264 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3272 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3275 struct btrfs_path *path;
3279 path = btrfs_alloc_path();
3284 ret = check_committed_ref(root, path, objectid,
3286 if (ret && ret != -ENOENT)
3289 ret2 = check_delayed_ref(root, path, objectid,
3291 } while (ret2 == -EAGAIN);
3293 if (ret2 && ret2 != -ENOENT) {
3298 if (ret != -ENOENT || ret2 != -ENOENT)
3301 btrfs_free_path(path);
3302 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3307 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3308 struct btrfs_root *root,
3309 struct extent_buffer *buf,
3310 int full_backref, int inc)
3312 struct btrfs_fs_info *fs_info = root->fs_info;
3318 struct btrfs_key key;
3319 struct btrfs_file_extent_item *fi;
3323 int (*process_func)(struct btrfs_trans_handle *,
3324 struct btrfs_root *,
3325 u64, u64, u64, u64, u64, u64);
3328 if (btrfs_is_testing(fs_info))
3331 ref_root = btrfs_header_owner(buf);
3332 nritems = btrfs_header_nritems(buf);
3333 level = btrfs_header_level(buf);
3335 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3339 process_func = btrfs_inc_extent_ref;
3341 process_func = btrfs_free_extent;
3344 parent = buf->start;
3348 for (i = 0; i < nritems; i++) {
3350 btrfs_item_key_to_cpu(buf, &key, i);
3351 if (key.type != BTRFS_EXTENT_DATA_KEY)
3353 fi = btrfs_item_ptr(buf, i,
3354 struct btrfs_file_extent_item);
3355 if (btrfs_file_extent_type(buf, fi) ==
3356 BTRFS_FILE_EXTENT_INLINE)
3358 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3362 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3363 key.offset -= btrfs_file_extent_offset(buf, fi);
3364 ret = process_func(trans, root, bytenr, num_bytes,
3365 parent, ref_root, key.objectid,
3370 bytenr = btrfs_node_blockptr(buf, i);
3371 num_bytes = fs_info->nodesize;
3372 ret = process_func(trans, root, bytenr, num_bytes,
3373 parent, ref_root, level - 1, 0);
3383 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3384 struct extent_buffer *buf, int full_backref)
3386 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3389 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3390 struct extent_buffer *buf, int full_backref)
3392 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3395 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3396 struct btrfs_fs_info *fs_info,
3397 struct btrfs_path *path,
3398 struct btrfs_block_group_cache *cache)
3401 struct btrfs_root *extent_root = fs_info->extent_root;
3403 struct extent_buffer *leaf;
3405 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3412 leaf = path->nodes[0];
3413 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3414 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3415 btrfs_mark_buffer_dirty(leaf);
3417 btrfs_release_path(path);
3422 static struct btrfs_block_group_cache *
3423 next_block_group(struct btrfs_fs_info *fs_info,
3424 struct btrfs_block_group_cache *cache)
3426 struct rb_node *node;
3428 spin_lock(&fs_info->block_group_cache_lock);
3430 /* If our block group was removed, we need a full search. */
3431 if (RB_EMPTY_NODE(&cache->cache_node)) {
3432 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3434 spin_unlock(&fs_info->block_group_cache_lock);
3435 btrfs_put_block_group(cache);
3436 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3438 node = rb_next(&cache->cache_node);
3439 btrfs_put_block_group(cache);
3441 cache = rb_entry(node, struct btrfs_block_group_cache,
3443 btrfs_get_block_group(cache);
3446 spin_unlock(&fs_info->block_group_cache_lock);
3450 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3451 struct btrfs_trans_handle *trans,
3452 struct btrfs_path *path)
3454 struct btrfs_fs_info *fs_info = block_group->fs_info;
3455 struct btrfs_root *root = fs_info->tree_root;
3456 struct inode *inode = NULL;
3457 struct extent_changeset *data_reserved = NULL;
3459 int dcs = BTRFS_DC_ERROR;
3465 * If this block group is smaller than 100 megs don't bother caching the
3468 if (block_group->key.offset < (100 * SZ_1M)) {
3469 spin_lock(&block_group->lock);
3470 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3471 spin_unlock(&block_group->lock);
3478 inode = lookup_free_space_inode(fs_info, block_group, path);
3479 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3480 ret = PTR_ERR(inode);
3481 btrfs_release_path(path);
3485 if (IS_ERR(inode)) {
3489 if (block_group->ro)
3492 ret = create_free_space_inode(fs_info, trans, block_group,
3500 * We want to set the generation to 0, that way if anything goes wrong
3501 * from here on out we know not to trust this cache when we load up next
3504 BTRFS_I(inode)->generation = 0;
3505 ret = btrfs_update_inode(trans, root, inode);
3508 * So theoretically we could recover from this, simply set the
3509 * super cache generation to 0 so we know to invalidate the
3510 * cache, but then we'd have to keep track of the block groups
3511 * that fail this way so we know we _have_ to reset this cache
3512 * before the next commit or risk reading stale cache. So to
3513 * limit our exposure to horrible edge cases lets just abort the
3514 * transaction, this only happens in really bad situations
3517 btrfs_abort_transaction(trans, ret);
3522 /* We've already setup this transaction, go ahead and exit */
3523 if (block_group->cache_generation == trans->transid &&
3524 i_size_read(inode)) {
3525 dcs = BTRFS_DC_SETUP;
3529 if (i_size_read(inode) > 0) {
3530 ret = btrfs_check_trunc_cache_free_space(fs_info,
3531 &fs_info->global_block_rsv);
3535 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3540 spin_lock(&block_group->lock);
3541 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3542 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3544 * don't bother trying to write stuff out _if_
3545 * a) we're not cached,
3546 * b) we're with nospace_cache mount option,
3547 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3549 dcs = BTRFS_DC_WRITTEN;
3550 spin_unlock(&block_group->lock);
3553 spin_unlock(&block_group->lock);
3556 * We hit an ENOSPC when setting up the cache in this transaction, just
3557 * skip doing the setup, we've already cleared the cache so we're safe.
3559 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3565 * Try to preallocate enough space based on how big the block group is.
3566 * Keep in mind this has to include any pinned space which could end up
3567 * taking up quite a bit since it's not folded into the other space
3570 num_pages = div_u64(block_group->key.offset, SZ_256M);
3575 num_pages *= PAGE_SIZE;
3577 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3581 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3582 num_pages, num_pages,
3585 * Our cache requires contiguous chunks so that we don't modify a bunch
3586 * of metadata or split extents when writing the cache out, which means
3587 * we can enospc if we are heavily fragmented in addition to just normal
3588 * out of space conditions. So if we hit this just skip setting up any
3589 * other block groups for this transaction, maybe we'll unpin enough
3590 * space the next time around.
3593 dcs = BTRFS_DC_SETUP;
3594 else if (ret == -ENOSPC)
3595 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3600 btrfs_release_path(path);
3602 spin_lock(&block_group->lock);
3603 if (!ret && dcs == BTRFS_DC_SETUP)
3604 block_group->cache_generation = trans->transid;
3605 block_group->disk_cache_state = dcs;
3606 spin_unlock(&block_group->lock);
3608 extent_changeset_free(data_reserved);
3612 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3613 struct btrfs_fs_info *fs_info)
3615 struct btrfs_block_group_cache *cache, *tmp;
3616 struct btrfs_transaction *cur_trans = trans->transaction;
3617 struct btrfs_path *path;
3619 if (list_empty(&cur_trans->dirty_bgs) ||
3620 !btrfs_test_opt(fs_info, SPACE_CACHE))
3623 path = btrfs_alloc_path();
3627 /* Could add new block groups, use _safe just in case */
3628 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3630 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3631 cache_save_setup(cache, trans, path);
3634 btrfs_free_path(path);
3639 * transaction commit does final block group cache writeback during a
3640 * critical section where nothing is allowed to change the FS. This is
3641 * required in order for the cache to actually match the block group,
3642 * but can introduce a lot of latency into the commit.
3644 * So, btrfs_start_dirty_block_groups is here to kick off block group
3645 * cache IO. There's a chance we'll have to redo some of it if the
3646 * block group changes again during the commit, but it greatly reduces
3647 * the commit latency by getting rid of the easy block groups while
3648 * we're still allowing others to join the commit.
3650 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3652 struct btrfs_fs_info *fs_info = trans->fs_info;
3653 struct btrfs_block_group_cache *cache;
3654 struct btrfs_transaction *cur_trans = trans->transaction;
3657 struct btrfs_path *path = NULL;
3659 struct list_head *io = &cur_trans->io_bgs;
3660 int num_started = 0;
3663 spin_lock(&cur_trans->dirty_bgs_lock);
3664 if (list_empty(&cur_trans->dirty_bgs)) {
3665 spin_unlock(&cur_trans->dirty_bgs_lock);
3668 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3669 spin_unlock(&cur_trans->dirty_bgs_lock);
3673 * make sure all the block groups on our dirty list actually
3676 btrfs_create_pending_block_groups(trans);
3679 path = btrfs_alloc_path();
3685 * cache_write_mutex is here only to save us from balance or automatic
3686 * removal of empty block groups deleting this block group while we are
3687 * writing out the cache
3689 mutex_lock(&trans->transaction->cache_write_mutex);
3690 while (!list_empty(&dirty)) {
3691 cache = list_first_entry(&dirty,
3692 struct btrfs_block_group_cache,
3695 * this can happen if something re-dirties a block
3696 * group that is already under IO. Just wait for it to
3697 * finish and then do it all again
3699 if (!list_empty(&cache->io_list)) {
3700 list_del_init(&cache->io_list);
3701 btrfs_wait_cache_io(trans, cache, path);
3702 btrfs_put_block_group(cache);
3707 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3708 * if it should update the cache_state. Don't delete
3709 * until after we wait.
3711 * Since we're not running in the commit critical section
3712 * we need the dirty_bgs_lock to protect from update_block_group
3714 spin_lock(&cur_trans->dirty_bgs_lock);
3715 list_del_init(&cache->dirty_list);
3716 spin_unlock(&cur_trans->dirty_bgs_lock);
3720 cache_save_setup(cache, trans, path);
3722 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3723 cache->io_ctl.inode = NULL;
3724 ret = btrfs_write_out_cache(fs_info, trans,
3726 if (ret == 0 && cache->io_ctl.inode) {
3731 * The cache_write_mutex is protecting the
3732 * io_list, also refer to the definition of
3733 * btrfs_transaction::io_bgs for more details
3735 list_add_tail(&cache->io_list, io);
3738 * if we failed to write the cache, the
3739 * generation will be bad and life goes on
3745 ret = write_one_cache_group(trans, fs_info,
3748 * Our block group might still be attached to the list
3749 * of new block groups in the transaction handle of some
3750 * other task (struct btrfs_trans_handle->new_bgs). This
3751 * means its block group item isn't yet in the extent
3752 * tree. If this happens ignore the error, as we will
3753 * try again later in the critical section of the
3754 * transaction commit.
3756 if (ret == -ENOENT) {
3758 spin_lock(&cur_trans->dirty_bgs_lock);
3759 if (list_empty(&cache->dirty_list)) {
3760 list_add_tail(&cache->dirty_list,
3761 &cur_trans->dirty_bgs);
3762 btrfs_get_block_group(cache);
3764 spin_unlock(&cur_trans->dirty_bgs_lock);
3766 btrfs_abort_transaction(trans, ret);
3770 /* if its not on the io list, we need to put the block group */
3772 btrfs_put_block_group(cache);
3778 * Avoid blocking other tasks for too long. It might even save
3779 * us from writing caches for block groups that are going to be
3782 mutex_unlock(&trans->transaction->cache_write_mutex);
3783 mutex_lock(&trans->transaction->cache_write_mutex);
3785 mutex_unlock(&trans->transaction->cache_write_mutex);
3788 * go through delayed refs for all the stuff we've just kicked off
3789 * and then loop back (just once)
3791 ret = btrfs_run_delayed_refs(trans, 0);
3792 if (!ret && loops == 0) {
3794 spin_lock(&cur_trans->dirty_bgs_lock);
3795 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3797 * dirty_bgs_lock protects us from concurrent block group
3798 * deletes too (not just cache_write_mutex).
3800 if (!list_empty(&dirty)) {
3801 spin_unlock(&cur_trans->dirty_bgs_lock);
3804 spin_unlock(&cur_trans->dirty_bgs_lock);
3805 } else if (ret < 0) {
3806 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3809 btrfs_free_path(path);
3813 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3814 struct btrfs_fs_info *fs_info)
3816 struct btrfs_block_group_cache *cache;
3817 struct btrfs_transaction *cur_trans = trans->transaction;
3820 struct btrfs_path *path;
3821 struct list_head *io = &cur_trans->io_bgs;
3822 int num_started = 0;
3824 path = btrfs_alloc_path();
3829 * Even though we are in the critical section of the transaction commit,
3830 * we can still have concurrent tasks adding elements to this
3831 * transaction's list of dirty block groups. These tasks correspond to
3832 * endio free space workers started when writeback finishes for a
3833 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3834 * allocate new block groups as a result of COWing nodes of the root
3835 * tree when updating the free space inode. The writeback for the space
3836 * caches is triggered by an earlier call to
3837 * btrfs_start_dirty_block_groups() and iterations of the following
3839 * Also we want to do the cache_save_setup first and then run the
3840 * delayed refs to make sure we have the best chance at doing this all
3843 spin_lock(&cur_trans->dirty_bgs_lock);
3844 while (!list_empty(&cur_trans->dirty_bgs)) {
3845 cache = list_first_entry(&cur_trans->dirty_bgs,
3846 struct btrfs_block_group_cache,
3850 * this can happen if cache_save_setup re-dirties a block
3851 * group that is already under IO. Just wait for it to
3852 * finish and then do it all again
3854 if (!list_empty(&cache->io_list)) {
3855 spin_unlock(&cur_trans->dirty_bgs_lock);
3856 list_del_init(&cache->io_list);
3857 btrfs_wait_cache_io(trans, cache, path);
3858 btrfs_put_block_group(cache);
3859 spin_lock(&cur_trans->dirty_bgs_lock);
3863 * don't remove from the dirty list until after we've waited
3866 list_del_init(&cache->dirty_list);
3867 spin_unlock(&cur_trans->dirty_bgs_lock);
3870 cache_save_setup(cache, trans, path);
3873 ret = btrfs_run_delayed_refs(trans,
3874 (unsigned long) -1);
3876 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3877 cache->io_ctl.inode = NULL;
3878 ret = btrfs_write_out_cache(fs_info, trans,
3880 if (ret == 0 && cache->io_ctl.inode) {
3883 list_add_tail(&cache->io_list, io);
3886 * if we failed to write the cache, the
3887 * generation will be bad and life goes on
3893 ret = write_one_cache_group(trans, fs_info,
3896 * One of the free space endio workers might have
3897 * created a new block group while updating a free space
3898 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3899 * and hasn't released its transaction handle yet, in
3900 * which case the new block group is still attached to
3901 * its transaction handle and its creation has not
3902 * finished yet (no block group item in the extent tree
3903 * yet, etc). If this is the case, wait for all free
3904 * space endio workers to finish and retry. This is a
3905 * a very rare case so no need for a more efficient and
3908 if (ret == -ENOENT) {
3909 wait_event(cur_trans->writer_wait,
3910 atomic_read(&cur_trans->num_writers) == 1);
3911 ret = write_one_cache_group(trans, fs_info,
3915 btrfs_abort_transaction(trans, ret);
3918 /* if its not on the io list, we need to put the block group */
3920 btrfs_put_block_group(cache);
3921 spin_lock(&cur_trans->dirty_bgs_lock);
3923 spin_unlock(&cur_trans->dirty_bgs_lock);
3926 * Refer to the definition of io_bgs member for details why it's safe
3927 * to use it without any locking
3929 while (!list_empty(io)) {
3930 cache = list_first_entry(io, struct btrfs_block_group_cache,
3932 list_del_init(&cache->io_list);
3933 btrfs_wait_cache_io(trans, cache, path);
3934 btrfs_put_block_group(cache);
3937 btrfs_free_path(path);
3941 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3943 struct btrfs_block_group_cache *block_group;
3946 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3947 if (!block_group || block_group->ro)
3950 btrfs_put_block_group(block_group);
3954 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3956 struct btrfs_block_group_cache *bg;
3959 bg = btrfs_lookup_block_group(fs_info, bytenr);
3963 spin_lock(&bg->lock);
3967 atomic_inc(&bg->nocow_writers);
3968 spin_unlock(&bg->lock);
3970 /* no put on block group, done by btrfs_dec_nocow_writers */
3972 btrfs_put_block_group(bg);
3978 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3980 struct btrfs_block_group_cache *bg;
3982 bg = btrfs_lookup_block_group(fs_info, bytenr);
3984 if (atomic_dec_and_test(&bg->nocow_writers))
3985 wake_up_atomic_t(&bg->nocow_writers);
3987 * Once for our lookup and once for the lookup done by a previous call
3988 * to btrfs_inc_nocow_writers()
3990 btrfs_put_block_group(bg);
3991 btrfs_put_block_group(bg);
3994 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3996 wait_on_atomic_t(&bg->nocow_writers, atomic_t_wait,
3997 TASK_UNINTERRUPTIBLE);
4000 static const char *alloc_name(u64 flags)
4003 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4005 case BTRFS_BLOCK_GROUP_METADATA:
4007 case BTRFS_BLOCK_GROUP_DATA:
4009 case BTRFS_BLOCK_GROUP_SYSTEM:
4013 return "invalid-combination";
4017 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4018 struct btrfs_space_info **new)
4021 struct btrfs_space_info *space_info;
4025 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4029 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4036 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4037 INIT_LIST_HEAD(&space_info->block_groups[i]);
4038 init_rwsem(&space_info->groups_sem);
4039 spin_lock_init(&space_info->lock);
4040 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4041 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4042 init_waitqueue_head(&space_info->wait);
4043 INIT_LIST_HEAD(&space_info->ro_bgs);
4044 INIT_LIST_HEAD(&space_info->tickets);
4045 INIT_LIST_HEAD(&space_info->priority_tickets);
4047 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4048 info->space_info_kobj, "%s",
4049 alloc_name(space_info->flags));
4051 percpu_counter_destroy(&space_info->total_bytes_pinned);
4057 list_add_rcu(&space_info->list, &info->space_info);
4058 if (flags & BTRFS_BLOCK_GROUP_DATA)
4059 info->data_sinfo = space_info;
4064 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4065 u64 total_bytes, u64 bytes_used,
4067 struct btrfs_space_info **space_info)
4069 struct btrfs_space_info *found;
4072 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4073 BTRFS_BLOCK_GROUP_RAID10))
4078 found = __find_space_info(info, flags);
4080 spin_lock(&found->lock);
4081 found->total_bytes += total_bytes;
4082 found->disk_total += total_bytes * factor;
4083 found->bytes_used += bytes_used;
4084 found->disk_used += bytes_used * factor;
4085 found->bytes_readonly += bytes_readonly;
4086 if (total_bytes > 0)
4088 space_info_add_new_bytes(info, found, total_bytes -
4089 bytes_used - bytes_readonly);
4090 spin_unlock(&found->lock);
4091 *space_info = found;
4094 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4096 u64 extra_flags = chunk_to_extended(flags) &
4097 BTRFS_EXTENDED_PROFILE_MASK;
4099 write_seqlock(&fs_info->profiles_lock);
4100 if (flags & BTRFS_BLOCK_GROUP_DATA)
4101 fs_info->avail_data_alloc_bits |= extra_flags;
4102 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4103 fs_info->avail_metadata_alloc_bits |= extra_flags;
4104 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4105 fs_info->avail_system_alloc_bits |= extra_flags;
4106 write_sequnlock(&fs_info->profiles_lock);
4110 * returns target flags in extended format or 0 if restripe for this
4111 * chunk_type is not in progress
4113 * should be called with either volume_mutex or balance_lock held
4115 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4117 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4123 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4124 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4125 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4126 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4127 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4128 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4129 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4130 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4131 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4138 * @flags: available profiles in extended format (see ctree.h)
4140 * Returns reduced profile in chunk format. If profile changing is in
4141 * progress (either running or paused) picks the target profile (if it's
4142 * already available), otherwise falls back to plain reducing.
4144 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4146 u64 num_devices = fs_info->fs_devices->rw_devices;
4152 * see if restripe for this chunk_type is in progress, if so
4153 * try to reduce to the target profile
4155 spin_lock(&fs_info->balance_lock);
4156 target = get_restripe_target(fs_info, flags);
4158 /* pick target profile only if it's already available */
4159 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4160 spin_unlock(&fs_info->balance_lock);
4161 return extended_to_chunk(target);
4164 spin_unlock(&fs_info->balance_lock);
4166 /* First, mask out the RAID levels which aren't possible */
4167 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4168 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4169 allowed |= btrfs_raid_group[raid_type];
4173 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4174 allowed = BTRFS_BLOCK_GROUP_RAID6;
4175 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4176 allowed = BTRFS_BLOCK_GROUP_RAID5;
4177 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4178 allowed = BTRFS_BLOCK_GROUP_RAID10;
4179 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4180 allowed = BTRFS_BLOCK_GROUP_RAID1;
4181 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4182 allowed = BTRFS_BLOCK_GROUP_RAID0;
4184 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4186 return extended_to_chunk(flags | allowed);
4189 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4196 seq = read_seqbegin(&fs_info->profiles_lock);
4198 if (flags & BTRFS_BLOCK_GROUP_DATA)
4199 flags |= fs_info->avail_data_alloc_bits;
4200 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4201 flags |= fs_info->avail_system_alloc_bits;
4202 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4203 flags |= fs_info->avail_metadata_alloc_bits;
4204 } while (read_seqretry(&fs_info->profiles_lock, seq));
4206 return btrfs_reduce_alloc_profile(fs_info, flags);
4209 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4211 struct btrfs_fs_info *fs_info = root->fs_info;
4216 flags = BTRFS_BLOCK_GROUP_DATA;
4217 else if (root == fs_info->chunk_root)
4218 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4220 flags = BTRFS_BLOCK_GROUP_METADATA;
4222 ret = get_alloc_profile(fs_info, flags);
4226 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4228 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4231 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4233 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4236 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4238 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4241 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4242 bool may_use_included)
4245 return s_info->bytes_used + s_info->bytes_reserved +
4246 s_info->bytes_pinned + s_info->bytes_readonly +
4247 (may_use_included ? s_info->bytes_may_use : 0);
4250 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4252 struct btrfs_root *root = inode->root;
4253 struct btrfs_fs_info *fs_info = root->fs_info;
4254 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4257 int need_commit = 2;
4258 int have_pinned_space;
4260 /* make sure bytes are sectorsize aligned */
4261 bytes = ALIGN(bytes, fs_info->sectorsize);
4263 if (btrfs_is_free_space_inode(inode)) {
4265 ASSERT(current->journal_info);
4269 /* make sure we have enough space to handle the data first */
4270 spin_lock(&data_sinfo->lock);
4271 used = btrfs_space_info_used(data_sinfo, true);
4273 if (used + bytes > data_sinfo->total_bytes) {
4274 struct btrfs_trans_handle *trans;
4277 * if we don't have enough free bytes in this space then we need
4278 * to alloc a new chunk.
4280 if (!data_sinfo->full) {
4283 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4284 spin_unlock(&data_sinfo->lock);
4286 alloc_target = btrfs_data_alloc_profile(fs_info);
4288 * It is ugly that we don't call nolock join
4289 * transaction for the free space inode case here.
4290 * But it is safe because we only do the data space
4291 * reservation for the free space cache in the
4292 * transaction context, the common join transaction
4293 * just increase the counter of the current transaction
4294 * handler, doesn't try to acquire the trans_lock of
4297 trans = btrfs_join_transaction(root);
4299 return PTR_ERR(trans);
4301 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4302 CHUNK_ALLOC_NO_FORCE);
4303 btrfs_end_transaction(trans);
4308 have_pinned_space = 1;
4317 * If we don't have enough pinned space to deal with this
4318 * allocation, and no removed chunk in current transaction,
4319 * don't bother committing the transaction.
4321 have_pinned_space = percpu_counter_compare(
4322 &data_sinfo->total_bytes_pinned,
4323 used + bytes - data_sinfo->total_bytes);
4324 spin_unlock(&data_sinfo->lock);
4326 /* commit the current transaction and try again */
4331 if (need_commit > 0) {
4332 btrfs_start_delalloc_roots(fs_info, 0, -1);
4333 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4337 trans = btrfs_join_transaction(root);
4339 return PTR_ERR(trans);
4340 if (have_pinned_space >= 0 ||
4341 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4342 &trans->transaction->flags) ||
4344 ret = btrfs_commit_transaction(trans);
4348 * The cleaner kthread might still be doing iput
4349 * operations. Wait for it to finish so that
4350 * more space is released.
4352 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4353 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4356 btrfs_end_transaction(trans);
4360 trace_btrfs_space_reservation(fs_info,
4361 "space_info:enospc",
4362 data_sinfo->flags, bytes, 1);
4365 data_sinfo->bytes_may_use += bytes;
4366 trace_btrfs_space_reservation(fs_info, "space_info",
4367 data_sinfo->flags, bytes, 1);
4368 spin_unlock(&data_sinfo->lock);
4373 int btrfs_check_data_free_space(struct inode *inode,
4374 struct extent_changeset **reserved, u64 start, u64 len)
4376 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4379 /* align the range */
4380 len = round_up(start + len, fs_info->sectorsize) -
4381 round_down(start, fs_info->sectorsize);
4382 start = round_down(start, fs_info->sectorsize);
4384 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4388 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4389 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4391 btrfs_free_reserved_data_space_noquota(inode, start, len);
4398 * Called if we need to clear a data reservation for this inode
4399 * Normally in a error case.
4401 * This one will *NOT* use accurate qgroup reserved space API, just for case
4402 * which we can't sleep and is sure it won't affect qgroup reserved space.
4403 * Like clear_bit_hook().
4405 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4408 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4409 struct btrfs_space_info *data_sinfo;
4411 /* Make sure the range is aligned to sectorsize */
4412 len = round_up(start + len, fs_info->sectorsize) -
4413 round_down(start, fs_info->sectorsize);
4414 start = round_down(start, fs_info->sectorsize);
4416 data_sinfo = fs_info->data_sinfo;
4417 spin_lock(&data_sinfo->lock);
4418 if (WARN_ON(data_sinfo->bytes_may_use < len))
4419 data_sinfo->bytes_may_use = 0;
4421 data_sinfo->bytes_may_use -= len;
4422 trace_btrfs_space_reservation(fs_info, "space_info",
4423 data_sinfo->flags, len, 0);
4424 spin_unlock(&data_sinfo->lock);
4428 * Called if we need to clear a data reservation for this inode
4429 * Normally in a error case.
4431 * This one will handle the per-inode data rsv map for accurate reserved
4434 void btrfs_free_reserved_data_space(struct inode *inode,
4435 struct extent_changeset *reserved, u64 start, u64 len)
4437 struct btrfs_root *root = BTRFS_I(inode)->root;
4439 /* Make sure the range is aligned to sectorsize */
4440 len = round_up(start + len, root->fs_info->sectorsize) -
4441 round_down(start, root->fs_info->sectorsize);
4442 start = round_down(start, root->fs_info->sectorsize);
4444 btrfs_free_reserved_data_space_noquota(inode, start, len);
4445 btrfs_qgroup_free_data(inode, reserved, start, len);
4448 static void force_metadata_allocation(struct btrfs_fs_info *info)
4450 struct list_head *head = &info->space_info;
4451 struct btrfs_space_info *found;
4454 list_for_each_entry_rcu(found, head, list) {
4455 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4456 found->force_alloc = CHUNK_ALLOC_FORCE;
4461 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4463 return (global->size << 1);
4466 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4467 struct btrfs_space_info *sinfo, int force)
4469 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4470 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4473 if (force == CHUNK_ALLOC_FORCE)
4477 * We need to take into account the global rsv because for all intents
4478 * and purposes it's used space. Don't worry about locking the
4479 * global_rsv, it doesn't change except when the transaction commits.
4481 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4482 bytes_used += calc_global_rsv_need_space(global_rsv);
4485 * in limited mode, we want to have some free space up to
4486 * about 1% of the FS size.
4488 if (force == CHUNK_ALLOC_LIMITED) {
4489 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4490 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4492 if (sinfo->total_bytes - bytes_used < thresh)
4496 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4501 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4505 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4506 BTRFS_BLOCK_GROUP_RAID0 |
4507 BTRFS_BLOCK_GROUP_RAID5 |
4508 BTRFS_BLOCK_GROUP_RAID6))
4509 num_dev = fs_info->fs_devices->rw_devices;
4510 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4513 num_dev = 1; /* DUP or single */
4519 * If @is_allocation is true, reserve space in the system space info necessary
4520 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4523 void check_system_chunk(struct btrfs_trans_handle *trans,
4524 struct btrfs_fs_info *fs_info, u64 type)
4526 struct btrfs_space_info *info;
4533 * Needed because we can end up allocating a system chunk and for an
4534 * atomic and race free space reservation in the chunk block reserve.
4536 lockdep_assert_held(&fs_info->chunk_mutex);
4538 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4539 spin_lock(&info->lock);
4540 left = info->total_bytes - btrfs_space_info_used(info, true);
4541 spin_unlock(&info->lock);
4543 num_devs = get_profile_num_devs(fs_info, type);
4545 /* num_devs device items to update and 1 chunk item to add or remove */
4546 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4547 btrfs_calc_trans_metadata_size(fs_info, 1);
4549 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4550 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4551 left, thresh, type);
4552 dump_space_info(fs_info, info, 0, 0);
4555 if (left < thresh) {
4556 u64 flags = btrfs_system_alloc_profile(fs_info);
4559 * Ignore failure to create system chunk. We might end up not
4560 * needing it, as we might not need to COW all nodes/leafs from
4561 * the paths we visit in the chunk tree (they were already COWed
4562 * or created in the current transaction for example).
4564 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4568 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4569 &fs_info->chunk_block_rsv,
4570 thresh, BTRFS_RESERVE_NO_FLUSH);
4572 trans->chunk_bytes_reserved += thresh;
4577 * If force is CHUNK_ALLOC_FORCE:
4578 * - return 1 if it successfully allocates a chunk,
4579 * - return errors including -ENOSPC otherwise.
4580 * If force is NOT CHUNK_ALLOC_FORCE:
4581 * - return 0 if it doesn't need to allocate a new chunk,
4582 * - return 1 if it successfully allocates a chunk,
4583 * - return errors including -ENOSPC otherwise.
4585 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4586 struct btrfs_fs_info *fs_info, u64 flags, int force)
4588 struct btrfs_space_info *space_info;
4589 int wait_for_alloc = 0;
4592 /* Don't re-enter if we're already allocating a chunk */
4593 if (trans->allocating_chunk)
4596 space_info = __find_space_info(fs_info, flags);
4600 spin_lock(&space_info->lock);
4601 if (force < space_info->force_alloc)
4602 force = space_info->force_alloc;
4603 if (space_info->full) {
4604 if (should_alloc_chunk(fs_info, space_info, force))
4608 spin_unlock(&space_info->lock);
4612 if (!should_alloc_chunk(fs_info, space_info, force)) {
4613 spin_unlock(&space_info->lock);
4615 } else if (space_info->chunk_alloc) {
4618 space_info->chunk_alloc = 1;
4621 spin_unlock(&space_info->lock);
4623 mutex_lock(&fs_info->chunk_mutex);
4626 * The chunk_mutex is held throughout the entirety of a chunk
4627 * allocation, so once we've acquired the chunk_mutex we know that the
4628 * other guy is done and we need to recheck and see if we should
4631 if (wait_for_alloc) {
4632 mutex_unlock(&fs_info->chunk_mutex);
4638 trans->allocating_chunk = true;
4641 * If we have mixed data/metadata chunks we want to make sure we keep
4642 * allocating mixed chunks instead of individual chunks.
4644 if (btrfs_mixed_space_info(space_info))
4645 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4648 * if we're doing a data chunk, go ahead and make sure that
4649 * we keep a reasonable number of metadata chunks allocated in the
4652 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4653 fs_info->data_chunk_allocations++;
4654 if (!(fs_info->data_chunk_allocations %
4655 fs_info->metadata_ratio))
4656 force_metadata_allocation(fs_info);
4660 * Check if we have enough space in SYSTEM chunk because we may need
4661 * to update devices.
4663 check_system_chunk(trans, fs_info, flags);
4665 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4666 trans->allocating_chunk = false;
4668 spin_lock(&space_info->lock);
4669 if (ret < 0 && ret != -ENOSPC)
4672 space_info->full = 1;
4676 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4678 space_info->chunk_alloc = 0;
4679 spin_unlock(&space_info->lock);
4680 mutex_unlock(&fs_info->chunk_mutex);
4682 * When we allocate a new chunk we reserve space in the chunk block
4683 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4684 * add new nodes/leafs to it if we end up needing to do it when
4685 * inserting the chunk item and updating device items as part of the
4686 * second phase of chunk allocation, performed by
4687 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4688 * large number of new block groups to create in our transaction
4689 * handle's new_bgs list to avoid exhausting the chunk block reserve
4690 * in extreme cases - like having a single transaction create many new
4691 * block groups when starting to write out the free space caches of all
4692 * the block groups that were made dirty during the lifetime of the
4695 if (trans->can_flush_pending_bgs &&
4696 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4697 btrfs_create_pending_block_groups(trans);
4698 btrfs_trans_release_chunk_metadata(trans);
4703 static int can_overcommit(struct btrfs_fs_info *fs_info,
4704 struct btrfs_space_info *space_info, u64 bytes,
4705 enum btrfs_reserve_flush_enum flush,
4708 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4714 /* Don't overcommit when in mixed mode. */
4715 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4719 profile = btrfs_system_alloc_profile(fs_info);
4721 profile = btrfs_metadata_alloc_profile(fs_info);
4723 used = btrfs_space_info_used(space_info, false);
4726 * We only want to allow over committing if we have lots of actual space
4727 * free, but if we don't have enough space to handle the global reserve
4728 * space then we could end up having a real enospc problem when trying
4729 * to allocate a chunk or some other such important allocation.
4731 spin_lock(&global_rsv->lock);
4732 space_size = calc_global_rsv_need_space(global_rsv);
4733 spin_unlock(&global_rsv->lock);
4734 if (used + space_size >= space_info->total_bytes)
4737 used += space_info->bytes_may_use;
4739 avail = atomic64_read(&fs_info->free_chunk_space);
4742 * If we have dup, raid1 or raid10 then only half of the free
4743 * space is actually useable. For raid56, the space info used
4744 * doesn't include the parity drive, so we don't have to
4747 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4748 BTRFS_BLOCK_GROUP_RAID1 |
4749 BTRFS_BLOCK_GROUP_RAID10))
4753 * If we aren't flushing all things, let us overcommit up to
4754 * 1/2th of the space. If we can flush, don't let us overcommit
4755 * too much, let it overcommit up to 1/8 of the space.
4757 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4762 if (used + bytes < space_info->total_bytes + avail)
4767 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4768 unsigned long nr_pages, int nr_items)
4770 struct super_block *sb = fs_info->sb;
4772 if (down_read_trylock(&sb->s_umount)) {
4773 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4774 up_read(&sb->s_umount);
4777 * We needn't worry the filesystem going from r/w to r/o though
4778 * we don't acquire ->s_umount mutex, because the filesystem
4779 * should guarantee the delalloc inodes list be empty after
4780 * the filesystem is readonly(all dirty pages are written to
4783 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4784 if (!current->journal_info)
4785 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4789 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4795 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4796 nr = div64_u64(to_reclaim, bytes);
4802 #define EXTENT_SIZE_PER_ITEM SZ_256K
4805 * shrink metadata reservation for delalloc
4807 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4808 u64 orig, bool wait_ordered)
4810 struct btrfs_space_info *space_info;
4811 struct btrfs_trans_handle *trans;
4816 unsigned long nr_pages;
4819 /* Calc the number of the pages we need flush for space reservation */
4820 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4821 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4823 trans = (struct btrfs_trans_handle *)current->journal_info;
4824 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4826 delalloc_bytes = percpu_counter_sum_positive(
4827 &fs_info->delalloc_bytes);
4828 if (delalloc_bytes == 0) {
4832 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4837 while (delalloc_bytes && loops < 3) {
4838 max_reclaim = min(delalloc_bytes, to_reclaim);
4839 nr_pages = max_reclaim >> PAGE_SHIFT;
4840 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4842 * We need to wait for the async pages to actually start before
4845 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4849 if (max_reclaim <= nr_pages)
4852 max_reclaim -= nr_pages;
4854 wait_event(fs_info->async_submit_wait,
4855 atomic_read(&fs_info->async_delalloc_pages) <=
4858 spin_lock(&space_info->lock);
4859 if (list_empty(&space_info->tickets) &&
4860 list_empty(&space_info->priority_tickets)) {
4861 spin_unlock(&space_info->lock);
4864 spin_unlock(&space_info->lock);
4867 if (wait_ordered && !trans) {
4868 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4870 time_left = schedule_timeout_killable(1);
4874 delalloc_bytes = percpu_counter_sum_positive(
4875 &fs_info->delalloc_bytes);
4879 struct reserve_ticket {
4882 struct list_head list;
4883 wait_queue_head_t wait;
4887 * maybe_commit_transaction - possibly commit the transaction if its ok to
4888 * @root - the root we're allocating for
4889 * @bytes - the number of bytes we want to reserve
4890 * @force - force the commit
4892 * This will check to make sure that committing the transaction will actually
4893 * get us somewhere and then commit the transaction if it does. Otherwise it
4894 * will return -ENOSPC.
4896 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4897 struct btrfs_space_info *space_info)
4899 struct reserve_ticket *ticket = NULL;
4900 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4901 struct btrfs_trans_handle *trans;
4904 trans = (struct btrfs_trans_handle *)current->journal_info;
4908 spin_lock(&space_info->lock);
4909 if (!list_empty(&space_info->priority_tickets))
4910 ticket = list_first_entry(&space_info->priority_tickets,
4911 struct reserve_ticket, list);
4912 else if (!list_empty(&space_info->tickets))
4913 ticket = list_first_entry(&space_info->tickets,
4914 struct reserve_ticket, list);
4915 bytes = (ticket) ? ticket->bytes : 0;
4916 spin_unlock(&space_info->lock);
4921 /* See if there is enough pinned space to make this reservation */
4922 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4927 * See if there is some space in the delayed insertion reservation for
4930 if (space_info != delayed_rsv->space_info)
4933 spin_lock(&delayed_rsv->lock);
4934 if (delayed_rsv->size > bytes)
4937 bytes -= delayed_rsv->size;
4938 spin_unlock(&delayed_rsv->lock);
4940 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4946 trans = btrfs_join_transaction(fs_info->extent_root);
4950 return btrfs_commit_transaction(trans);
4954 * Try to flush some data based on policy set by @state. This is only advisory
4955 * and may fail for various reasons. The caller is supposed to examine the
4956 * state of @space_info to detect the outcome.
4958 static void flush_space(struct btrfs_fs_info *fs_info,
4959 struct btrfs_space_info *space_info, u64 num_bytes,
4962 struct btrfs_root *root = fs_info->extent_root;
4963 struct btrfs_trans_handle *trans;
4968 case FLUSH_DELAYED_ITEMS_NR:
4969 case FLUSH_DELAYED_ITEMS:
4970 if (state == FLUSH_DELAYED_ITEMS_NR)
4971 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4975 trans = btrfs_join_transaction(root);
4976 if (IS_ERR(trans)) {
4977 ret = PTR_ERR(trans);
4980 ret = btrfs_run_delayed_items_nr(trans, nr);
4981 btrfs_end_transaction(trans);
4983 case FLUSH_DELALLOC:
4984 case FLUSH_DELALLOC_WAIT:
4985 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4986 state == FLUSH_DELALLOC_WAIT);
4989 trans = btrfs_join_transaction(root);
4990 if (IS_ERR(trans)) {
4991 ret = PTR_ERR(trans);
4994 ret = do_chunk_alloc(trans, fs_info,
4995 btrfs_metadata_alloc_profile(fs_info),
4996 CHUNK_ALLOC_NO_FORCE);
4997 btrfs_end_transaction(trans);
4998 if (ret > 0 || ret == -ENOSPC)
5002 ret = may_commit_transaction(fs_info, space_info);
5009 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5015 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5016 struct btrfs_space_info *space_info,
5019 struct reserve_ticket *ticket;
5024 list_for_each_entry(ticket, &space_info->tickets, list)
5025 to_reclaim += ticket->bytes;
5026 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5027 to_reclaim += ticket->bytes;
5031 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5032 if (can_overcommit(fs_info, space_info, to_reclaim,
5033 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5036 used = btrfs_space_info_used(space_info, true);
5038 if (can_overcommit(fs_info, space_info, SZ_1M,
5039 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5040 expected = div_factor_fine(space_info->total_bytes, 95);
5042 expected = div_factor_fine(space_info->total_bytes, 90);
5044 if (used > expected)
5045 to_reclaim = used - expected;
5048 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5049 space_info->bytes_reserved);
5053 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5054 struct btrfs_space_info *space_info,
5055 u64 used, bool system_chunk)
5057 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5059 /* If we're just plain full then async reclaim just slows us down. */
5060 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5063 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5067 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5068 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5071 static void wake_all_tickets(struct list_head *head)
5073 struct reserve_ticket *ticket;
5075 while (!list_empty(head)) {
5076 ticket = list_first_entry(head, struct reserve_ticket, list);
5077 list_del_init(&ticket->list);
5078 ticket->error = -ENOSPC;
5079 wake_up(&ticket->wait);
5084 * This is for normal flushers, we can wait all goddamned day if we want to. We
5085 * will loop and continuously try to flush as long as we are making progress.
5086 * We count progress as clearing off tickets each time we have to loop.
5088 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5090 struct btrfs_fs_info *fs_info;
5091 struct btrfs_space_info *space_info;
5094 int commit_cycles = 0;
5095 u64 last_tickets_id;
5097 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5098 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5100 spin_lock(&space_info->lock);
5101 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5104 space_info->flush = 0;
5105 spin_unlock(&space_info->lock);
5108 last_tickets_id = space_info->tickets_id;
5109 spin_unlock(&space_info->lock);
5111 flush_state = FLUSH_DELAYED_ITEMS_NR;
5113 flush_space(fs_info, space_info, to_reclaim, flush_state);
5114 spin_lock(&space_info->lock);
5115 if (list_empty(&space_info->tickets)) {
5116 space_info->flush = 0;
5117 spin_unlock(&space_info->lock);
5120 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5123 if (last_tickets_id == space_info->tickets_id) {
5126 last_tickets_id = space_info->tickets_id;
5127 flush_state = FLUSH_DELAYED_ITEMS_NR;
5132 if (flush_state > COMMIT_TRANS) {
5134 if (commit_cycles > 2) {
5135 wake_all_tickets(&space_info->tickets);
5136 space_info->flush = 0;
5138 flush_state = FLUSH_DELAYED_ITEMS_NR;
5141 spin_unlock(&space_info->lock);
5142 } while (flush_state <= COMMIT_TRANS);
5145 void btrfs_init_async_reclaim_work(struct work_struct *work)
5147 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5150 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5151 struct btrfs_space_info *space_info,
5152 struct reserve_ticket *ticket)
5155 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5157 spin_lock(&space_info->lock);
5158 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5161 spin_unlock(&space_info->lock);
5164 spin_unlock(&space_info->lock);
5167 flush_space(fs_info, space_info, to_reclaim, flush_state);
5169 spin_lock(&space_info->lock);
5170 if (ticket->bytes == 0) {
5171 spin_unlock(&space_info->lock);
5174 spin_unlock(&space_info->lock);
5177 * Priority flushers can't wait on delalloc without
5180 if (flush_state == FLUSH_DELALLOC ||
5181 flush_state == FLUSH_DELALLOC_WAIT)
5182 flush_state = ALLOC_CHUNK;
5183 } while (flush_state < COMMIT_TRANS);
5186 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5187 struct btrfs_space_info *space_info,
5188 struct reserve_ticket *ticket, u64 orig_bytes)
5194 spin_lock(&space_info->lock);
5195 while (ticket->bytes > 0 && ticket->error == 0) {
5196 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5201 spin_unlock(&space_info->lock);
5205 finish_wait(&ticket->wait, &wait);
5206 spin_lock(&space_info->lock);
5209 ret = ticket->error;
5210 if (!list_empty(&ticket->list))
5211 list_del_init(&ticket->list);
5212 if (ticket->bytes && ticket->bytes < orig_bytes) {
5213 u64 num_bytes = orig_bytes - ticket->bytes;
5214 space_info->bytes_may_use -= num_bytes;
5215 trace_btrfs_space_reservation(fs_info, "space_info",
5216 space_info->flags, num_bytes, 0);
5218 spin_unlock(&space_info->lock);
5224 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5225 * @root - the root we're allocating for
5226 * @space_info - the space info we want to allocate from
5227 * @orig_bytes - the number of bytes we want
5228 * @flush - whether or not we can flush to make our reservation
5230 * This will reserve orig_bytes number of bytes from the space info associated
5231 * with the block_rsv. If there is not enough space it will make an attempt to
5232 * flush out space to make room. It will do this by flushing delalloc if
5233 * possible or committing the transaction. If flush is 0 then no attempts to
5234 * regain reservations will be made and this will fail if there is not enough
5237 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5238 struct btrfs_space_info *space_info,
5240 enum btrfs_reserve_flush_enum flush,
5243 struct reserve_ticket ticket;
5248 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5250 spin_lock(&space_info->lock);
5252 used = btrfs_space_info_used(space_info, true);
5255 * If we have enough space then hooray, make our reservation and carry
5256 * on. If not see if we can overcommit, and if we can, hooray carry on.
5257 * If not things get more complicated.
5259 if (used + orig_bytes <= space_info->total_bytes) {
5260 space_info->bytes_may_use += orig_bytes;
5261 trace_btrfs_space_reservation(fs_info, "space_info",
5262 space_info->flags, orig_bytes, 1);
5264 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5266 space_info->bytes_may_use += orig_bytes;
5267 trace_btrfs_space_reservation(fs_info, "space_info",
5268 space_info->flags, orig_bytes, 1);
5273 * If we couldn't make a reservation then setup our reservation ticket
5274 * and kick the async worker if it's not already running.
5276 * If we are a priority flusher then we just need to add our ticket to
5277 * the list and we will do our own flushing further down.
5279 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5280 ticket.bytes = orig_bytes;
5282 init_waitqueue_head(&ticket.wait);
5283 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5284 list_add_tail(&ticket.list, &space_info->tickets);
5285 if (!space_info->flush) {
5286 space_info->flush = 1;
5287 trace_btrfs_trigger_flush(fs_info,
5291 queue_work(system_unbound_wq,
5292 &fs_info->async_reclaim_work);
5295 list_add_tail(&ticket.list,
5296 &space_info->priority_tickets);
5298 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5301 * We will do the space reservation dance during log replay,
5302 * which means we won't have fs_info->fs_root set, so don't do
5303 * the async reclaim as we will panic.
5305 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5306 need_do_async_reclaim(fs_info, space_info,
5307 used, system_chunk) &&
5308 !work_busy(&fs_info->async_reclaim_work)) {
5309 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5310 orig_bytes, flush, "preempt");
5311 queue_work(system_unbound_wq,
5312 &fs_info->async_reclaim_work);
5315 spin_unlock(&space_info->lock);
5316 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5319 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5320 return wait_reserve_ticket(fs_info, space_info, &ticket,
5324 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5325 spin_lock(&space_info->lock);
5327 if (ticket.bytes < orig_bytes) {
5328 u64 num_bytes = orig_bytes - ticket.bytes;
5329 space_info->bytes_may_use -= num_bytes;
5330 trace_btrfs_space_reservation(fs_info, "space_info",
5335 list_del_init(&ticket.list);
5338 spin_unlock(&space_info->lock);
5339 ASSERT(list_empty(&ticket.list));
5344 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5345 * @root - the root we're allocating for
5346 * @block_rsv - the block_rsv we're allocating for
5347 * @orig_bytes - the number of bytes we want
5348 * @flush - whether or not we can flush to make our reservation
5350 * This will reserve orgi_bytes number of bytes from the space info associated
5351 * with the block_rsv. If there is not enough space it will make an attempt to
5352 * flush out space to make room. It will do this by flushing delalloc if
5353 * possible or committing the transaction. If flush is 0 then no attempts to
5354 * regain reservations will be made and this will fail if there is not enough
5357 static int reserve_metadata_bytes(struct btrfs_root *root,
5358 struct btrfs_block_rsv *block_rsv,
5360 enum btrfs_reserve_flush_enum flush)
5362 struct btrfs_fs_info *fs_info = root->fs_info;
5363 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5365 bool system_chunk = (root == fs_info->chunk_root);
5367 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5368 orig_bytes, flush, system_chunk);
5369 if (ret == -ENOSPC &&
5370 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5371 if (block_rsv != global_rsv &&
5372 !block_rsv_use_bytes(global_rsv, orig_bytes))
5375 if (ret == -ENOSPC) {
5376 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5377 block_rsv->space_info->flags,
5380 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5381 dump_space_info(fs_info, block_rsv->space_info,
5387 static struct btrfs_block_rsv *get_block_rsv(
5388 const struct btrfs_trans_handle *trans,
5389 const struct btrfs_root *root)
5391 struct btrfs_fs_info *fs_info = root->fs_info;
5392 struct btrfs_block_rsv *block_rsv = NULL;
5394 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5395 (root == fs_info->csum_root && trans->adding_csums) ||
5396 (root == fs_info->uuid_root))
5397 block_rsv = trans->block_rsv;
5400 block_rsv = root->block_rsv;
5403 block_rsv = &fs_info->empty_block_rsv;
5408 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5412 spin_lock(&block_rsv->lock);
5413 if (block_rsv->reserved >= num_bytes) {
5414 block_rsv->reserved -= num_bytes;
5415 if (block_rsv->reserved < block_rsv->size)
5416 block_rsv->full = 0;
5419 spin_unlock(&block_rsv->lock);
5423 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5424 u64 num_bytes, int update_size)
5426 spin_lock(&block_rsv->lock);
5427 block_rsv->reserved += num_bytes;
5429 block_rsv->size += num_bytes;
5430 else if (block_rsv->reserved >= block_rsv->size)
5431 block_rsv->full = 1;
5432 spin_unlock(&block_rsv->lock);
5435 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5436 struct btrfs_block_rsv *dest, u64 num_bytes,
5439 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5442 if (global_rsv->space_info != dest->space_info)
5445 spin_lock(&global_rsv->lock);
5446 min_bytes = div_factor(global_rsv->size, min_factor);
5447 if (global_rsv->reserved < min_bytes + num_bytes) {
5448 spin_unlock(&global_rsv->lock);
5451 global_rsv->reserved -= num_bytes;
5452 if (global_rsv->reserved < global_rsv->size)
5453 global_rsv->full = 0;
5454 spin_unlock(&global_rsv->lock);
5456 block_rsv_add_bytes(dest, num_bytes, 1);
5461 * This is for space we already have accounted in space_info->bytes_may_use, so
5462 * basically when we're returning space from block_rsv's.
5464 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5465 struct btrfs_space_info *space_info,
5468 struct reserve_ticket *ticket;
5469 struct list_head *head;
5471 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5472 bool check_overcommit = false;
5474 spin_lock(&space_info->lock);
5475 head = &space_info->priority_tickets;
5478 * If we are over our limit then we need to check and see if we can
5479 * overcommit, and if we can't then we just need to free up our space
5480 * and not satisfy any requests.
5482 used = btrfs_space_info_used(space_info, true);
5483 if (used - num_bytes >= space_info->total_bytes)
5484 check_overcommit = true;
5486 while (!list_empty(head) && num_bytes) {
5487 ticket = list_first_entry(head, struct reserve_ticket,
5490 * We use 0 bytes because this space is already reserved, so
5491 * adding the ticket space would be a double count.
5493 if (check_overcommit &&
5494 !can_overcommit(fs_info, space_info, 0, flush, false))
5496 if (num_bytes >= ticket->bytes) {
5497 list_del_init(&ticket->list);
5498 num_bytes -= ticket->bytes;
5500 space_info->tickets_id++;
5501 wake_up(&ticket->wait);
5503 ticket->bytes -= num_bytes;
5508 if (num_bytes && head == &space_info->priority_tickets) {
5509 head = &space_info->tickets;
5510 flush = BTRFS_RESERVE_FLUSH_ALL;
5513 space_info->bytes_may_use -= num_bytes;
5514 trace_btrfs_space_reservation(fs_info, "space_info",
5515 space_info->flags, num_bytes, 0);
5516 spin_unlock(&space_info->lock);
5520 * This is for newly allocated space that isn't accounted in
5521 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5522 * we use this helper.
5524 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5525 struct btrfs_space_info *space_info,
5528 struct reserve_ticket *ticket;
5529 struct list_head *head = &space_info->priority_tickets;
5532 while (!list_empty(head) && num_bytes) {
5533 ticket = list_first_entry(head, struct reserve_ticket,
5535 if (num_bytes >= ticket->bytes) {
5536 trace_btrfs_space_reservation(fs_info, "space_info",
5539 list_del_init(&ticket->list);
5540 num_bytes -= ticket->bytes;
5541 space_info->bytes_may_use += ticket->bytes;
5543 space_info->tickets_id++;
5544 wake_up(&ticket->wait);
5546 trace_btrfs_space_reservation(fs_info, "space_info",
5549 space_info->bytes_may_use += num_bytes;
5550 ticket->bytes -= num_bytes;
5555 if (num_bytes && head == &space_info->priority_tickets) {
5556 head = &space_info->tickets;
5561 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5562 struct btrfs_block_rsv *block_rsv,
5563 struct btrfs_block_rsv *dest, u64 num_bytes)
5565 struct btrfs_space_info *space_info = block_rsv->space_info;
5568 spin_lock(&block_rsv->lock);
5569 if (num_bytes == (u64)-1)
5570 num_bytes = block_rsv->size;
5571 block_rsv->size -= num_bytes;
5572 if (block_rsv->reserved >= block_rsv->size) {
5573 num_bytes = block_rsv->reserved - block_rsv->size;
5574 block_rsv->reserved = block_rsv->size;
5575 block_rsv->full = 1;
5579 spin_unlock(&block_rsv->lock);
5582 if (num_bytes > 0) {
5584 spin_lock(&dest->lock);
5588 bytes_to_add = dest->size - dest->reserved;
5589 bytes_to_add = min(num_bytes, bytes_to_add);
5590 dest->reserved += bytes_to_add;
5591 if (dest->reserved >= dest->size)
5593 num_bytes -= bytes_to_add;
5595 spin_unlock(&dest->lock);
5598 space_info_add_old_bytes(fs_info, space_info,
5604 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5605 struct btrfs_block_rsv *dst, u64 num_bytes,
5610 ret = block_rsv_use_bytes(src, num_bytes);
5614 block_rsv_add_bytes(dst, num_bytes, update_size);
5618 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5620 memset(rsv, 0, sizeof(*rsv));
5621 spin_lock_init(&rsv->lock);
5625 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5626 struct btrfs_block_rsv *rsv,
5627 unsigned short type)
5629 btrfs_init_block_rsv(rsv, type);
5630 rsv->space_info = __find_space_info(fs_info,
5631 BTRFS_BLOCK_GROUP_METADATA);
5634 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5635 unsigned short type)
5637 struct btrfs_block_rsv *block_rsv;
5639 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5643 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5647 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5648 struct btrfs_block_rsv *rsv)
5652 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5656 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5661 int btrfs_block_rsv_add(struct btrfs_root *root,
5662 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5663 enum btrfs_reserve_flush_enum flush)
5670 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5672 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5679 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5687 spin_lock(&block_rsv->lock);
5688 num_bytes = div_factor(block_rsv->size, min_factor);
5689 if (block_rsv->reserved >= num_bytes)
5691 spin_unlock(&block_rsv->lock);
5696 int btrfs_block_rsv_refill(struct btrfs_root *root,
5697 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5698 enum btrfs_reserve_flush_enum flush)
5706 spin_lock(&block_rsv->lock);
5707 num_bytes = min_reserved;
5708 if (block_rsv->reserved >= num_bytes)
5711 num_bytes -= block_rsv->reserved;
5712 spin_unlock(&block_rsv->lock);
5717 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5719 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5727 * btrfs_inode_rsv_refill - refill the inode block rsv.
5728 * @inode - the inode we are refilling.
5729 * @flush - the flusing restriction.
5731 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5732 * block_rsv->size as the minimum size. We'll either refill the missing amount
5733 * or return if we already have enough space. This will also handle the resreve
5734 * tracepoint for the reserved amount.
5736 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5737 enum btrfs_reserve_flush_enum flush)
5739 struct btrfs_root *root = inode->root;
5740 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5744 spin_lock(&block_rsv->lock);
5745 if (block_rsv->reserved < block_rsv->size)
5746 num_bytes = block_rsv->size - block_rsv->reserved;
5747 spin_unlock(&block_rsv->lock);
5752 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5755 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5757 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5758 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5759 btrfs_ino(inode), num_bytes, 1);
5765 * btrfs_inode_rsv_release - release any excessive reservation.
5766 * @inode - the inode we need to release from.
5767 * @qgroup_free - free or convert qgroup meta.
5768 * Unlike normal operation, qgroup meta reservation needs to know if we are
5769 * freeing qgroup reservation or just converting it into per-trans. Normally
5770 * @qgroup_free is true for error handling, and false for normal release.
5772 * This is the same as btrfs_block_rsv_release, except that it handles the
5773 * tracepoint for the reservation.
5775 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5777 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5778 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5779 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5783 * Since we statically set the block_rsv->size we just want to say we
5784 * are releasing 0 bytes, and then we'll just get the reservation over
5787 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0);
5789 trace_btrfs_space_reservation(fs_info, "delalloc",
5790 btrfs_ino(inode), released, 0);
5792 btrfs_qgroup_free_meta_prealloc(inode->root, released);
5794 btrfs_qgroup_convert_reserved_meta(inode->root, released);
5797 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5798 struct btrfs_block_rsv *block_rsv,
5801 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5803 if (global_rsv == block_rsv ||
5804 block_rsv->space_info != global_rsv->space_info)
5806 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5809 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5811 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5812 struct btrfs_space_info *sinfo = block_rsv->space_info;
5816 * The global block rsv is based on the size of the extent tree, the
5817 * checksum tree and the root tree. If the fs is empty we want to set
5818 * it to a minimal amount for safety.
5820 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5821 btrfs_root_used(&fs_info->csum_root->root_item) +
5822 btrfs_root_used(&fs_info->tree_root->root_item);
5823 num_bytes = max_t(u64, num_bytes, SZ_16M);
5825 spin_lock(&sinfo->lock);
5826 spin_lock(&block_rsv->lock);
5828 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5830 if (block_rsv->reserved < block_rsv->size) {
5831 num_bytes = btrfs_space_info_used(sinfo, true);
5832 if (sinfo->total_bytes > num_bytes) {
5833 num_bytes = sinfo->total_bytes - num_bytes;
5834 num_bytes = min(num_bytes,
5835 block_rsv->size - block_rsv->reserved);
5836 block_rsv->reserved += num_bytes;
5837 sinfo->bytes_may_use += num_bytes;
5838 trace_btrfs_space_reservation(fs_info, "space_info",
5839 sinfo->flags, num_bytes,
5842 } else if (block_rsv->reserved > block_rsv->size) {
5843 num_bytes = block_rsv->reserved - block_rsv->size;
5844 sinfo->bytes_may_use -= num_bytes;
5845 trace_btrfs_space_reservation(fs_info, "space_info",
5846 sinfo->flags, num_bytes, 0);
5847 block_rsv->reserved = block_rsv->size;
5850 if (block_rsv->reserved == block_rsv->size)
5851 block_rsv->full = 1;
5853 block_rsv->full = 0;
5855 spin_unlock(&block_rsv->lock);
5856 spin_unlock(&sinfo->lock);
5859 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5861 struct btrfs_space_info *space_info;
5863 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5864 fs_info->chunk_block_rsv.space_info = space_info;
5866 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5867 fs_info->global_block_rsv.space_info = space_info;
5868 fs_info->trans_block_rsv.space_info = space_info;
5869 fs_info->empty_block_rsv.space_info = space_info;
5870 fs_info->delayed_block_rsv.space_info = space_info;
5872 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5873 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5874 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5875 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5876 if (fs_info->quota_root)
5877 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5878 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5880 update_global_block_rsv(fs_info);
5883 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5885 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5887 WARN_ON(fs_info->trans_block_rsv.size > 0);
5888 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5889 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5890 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5891 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5892 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5897 * To be called after all the new block groups attached to the transaction
5898 * handle have been created (btrfs_create_pending_block_groups()).
5900 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5902 struct btrfs_fs_info *fs_info = trans->fs_info;
5904 if (!trans->chunk_bytes_reserved)
5907 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5909 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5910 trans->chunk_bytes_reserved);
5911 trans->chunk_bytes_reserved = 0;
5914 /* Can only return 0 or -ENOSPC */
5915 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5916 struct btrfs_inode *inode)
5918 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5919 struct btrfs_root *root = inode->root;
5921 * We always use trans->block_rsv here as we will have reserved space
5922 * for our orphan when starting the transaction, using get_block_rsv()
5923 * here will sometimes make us choose the wrong block rsv as we could be
5924 * doing a reloc inode for a non refcounted root.
5926 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5927 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5930 * We need to hold space in order to delete our orphan item once we've
5931 * added it, so this takes the reservation so we can release it later
5932 * when we are truly done with the orphan item.
5934 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5936 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5938 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5941 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5943 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5944 struct btrfs_root *root = inode->root;
5945 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5947 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5949 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5953 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5954 * root: the root of the parent directory
5955 * rsv: block reservation
5956 * items: the number of items that we need do reservation
5957 * qgroup_reserved: used to return the reserved size in qgroup
5959 * This function is used to reserve the space for snapshot/subvolume
5960 * creation and deletion. Those operations are different with the
5961 * common file/directory operations, they change two fs/file trees
5962 * and root tree, the number of items that the qgroup reserves is
5963 * different with the free space reservation. So we can not use
5964 * the space reservation mechanism in start_transaction().
5966 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5967 struct btrfs_block_rsv *rsv,
5969 u64 *qgroup_reserved,
5970 bool use_global_rsv)
5974 struct btrfs_fs_info *fs_info = root->fs_info;
5975 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5977 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5978 /* One for parent inode, two for dir entries */
5979 num_bytes = 3 * fs_info->nodesize;
5980 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5987 *qgroup_reserved = num_bytes;
5989 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5990 rsv->space_info = __find_space_info(fs_info,
5991 BTRFS_BLOCK_GROUP_METADATA);
5992 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5993 BTRFS_RESERVE_FLUSH_ALL);
5995 if (ret == -ENOSPC && use_global_rsv)
5996 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5998 if (ret && *qgroup_reserved)
5999 btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved);
6004 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6005 struct btrfs_block_rsv *rsv)
6007 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6010 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6011 struct btrfs_inode *inode)
6013 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6014 u64 reserve_size = 0;
6016 unsigned outstanding_extents;
6018 lockdep_assert_held(&inode->lock);
6019 outstanding_extents = inode->outstanding_extents;
6020 if (outstanding_extents)
6021 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6022 outstanding_extents + 1);
6023 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6025 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6028 spin_lock(&block_rsv->lock);
6029 block_rsv->size = reserve_size;
6030 spin_unlock(&block_rsv->lock);
6033 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6035 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6036 unsigned nr_extents;
6037 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6039 bool delalloc_lock = true;
6041 /* If we are a free space inode we need to not flush since we will be in
6042 * the middle of a transaction commit. We also don't need the delalloc
6043 * mutex since we won't race with anybody. We need this mostly to make
6044 * lockdep shut its filthy mouth.
6046 * If we have a transaction open (can happen if we call truncate_block
6047 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6049 if (btrfs_is_free_space_inode(inode)) {
6050 flush = BTRFS_RESERVE_NO_FLUSH;
6051 delalloc_lock = false;
6053 if (current->journal_info)
6054 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6056 if (btrfs_transaction_in_commit(fs_info))
6057 schedule_timeout(1);
6061 mutex_lock(&inode->delalloc_mutex);
6063 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6065 /* Add our new extents and calculate the new rsv size. */
6066 spin_lock(&inode->lock);
6067 nr_extents = count_max_extents(num_bytes);
6068 btrfs_mod_outstanding_extents(inode, nr_extents);
6069 inode->csum_bytes += num_bytes;
6070 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6071 spin_unlock(&inode->lock);
6073 ret = btrfs_inode_rsv_refill(inode, flush);
6078 mutex_unlock(&inode->delalloc_mutex);
6082 spin_lock(&inode->lock);
6083 nr_extents = count_max_extents(num_bytes);
6084 btrfs_mod_outstanding_extents(inode, -nr_extents);
6085 inode->csum_bytes -= num_bytes;
6086 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6087 spin_unlock(&inode->lock);
6089 btrfs_inode_rsv_release(inode, true);
6091 mutex_unlock(&inode->delalloc_mutex);
6096 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6097 * @inode: the inode to release the reservation for.
6098 * @num_bytes: the number of bytes we are releasing.
6099 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6101 * This will release the metadata reservation for an inode. This can be called
6102 * once we complete IO for a given set of bytes to release their metadata
6103 * reservations, or on error for the same reason.
6105 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6108 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6110 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6111 spin_lock(&inode->lock);
6112 inode->csum_bytes -= num_bytes;
6113 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6114 spin_unlock(&inode->lock);
6116 if (btrfs_is_testing(fs_info))
6119 btrfs_inode_rsv_release(inode, qgroup_free);
6123 * btrfs_delalloc_release_extents - release our outstanding_extents
6124 * @inode: the inode to balance the reservation for.
6125 * @num_bytes: the number of bytes we originally reserved with
6126 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6128 * When we reserve space we increase outstanding_extents for the extents we may
6129 * add. Once we've set the range as delalloc or created our ordered extents we
6130 * have outstanding_extents to track the real usage, so we use this to free our
6131 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6132 * with btrfs_delalloc_reserve_metadata.
6134 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6137 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6138 unsigned num_extents;
6140 spin_lock(&inode->lock);
6141 num_extents = count_max_extents(num_bytes);
6142 btrfs_mod_outstanding_extents(inode, -num_extents);
6143 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6144 spin_unlock(&inode->lock);
6146 if (btrfs_is_testing(fs_info))
6149 btrfs_inode_rsv_release(inode, qgroup_free);
6153 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6155 * @inode: inode we're writing to
6156 * @start: start range we are writing to
6157 * @len: how long the range we are writing to
6158 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6159 * current reservation.
6161 * This will do the following things
6163 * o reserve space in data space info for num bytes
6164 * and reserve precious corresponding qgroup space
6165 * (Done in check_data_free_space)
6167 * o reserve space for metadata space, based on the number of outstanding
6168 * extents and how much csums will be needed
6169 * also reserve metadata space in a per root over-reserve method.
6170 * o add to the inodes->delalloc_bytes
6171 * o add it to the fs_info's delalloc inodes list.
6172 * (Above 3 all done in delalloc_reserve_metadata)
6174 * Return 0 for success
6175 * Return <0 for error(-ENOSPC or -EQUOT)
6177 int btrfs_delalloc_reserve_space(struct inode *inode,
6178 struct extent_changeset **reserved, u64 start, u64 len)
6182 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6185 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6187 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6192 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6193 * @inode: inode we're releasing space for
6194 * @start: start position of the space already reserved
6195 * @len: the len of the space already reserved
6196 * @release_bytes: the len of the space we consumed or didn't use
6198 * This function will release the metadata space that was not used and will
6199 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6200 * list if there are no delalloc bytes left.
6201 * Also it will handle the qgroup reserved space.
6203 void btrfs_delalloc_release_space(struct inode *inode,
6204 struct extent_changeset *reserved,
6205 u64 start, u64 len, bool qgroup_free)
6207 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6208 btrfs_free_reserved_data_space(inode, reserved, start, len);
6211 static int update_block_group(struct btrfs_trans_handle *trans,
6212 struct btrfs_fs_info *info, u64 bytenr,
6213 u64 num_bytes, int alloc)
6215 struct btrfs_block_group_cache *cache = NULL;
6216 u64 total = num_bytes;
6221 /* block accounting for super block */
6222 spin_lock(&info->delalloc_root_lock);
6223 old_val = btrfs_super_bytes_used(info->super_copy);
6225 old_val += num_bytes;
6227 old_val -= num_bytes;
6228 btrfs_set_super_bytes_used(info->super_copy, old_val);
6229 spin_unlock(&info->delalloc_root_lock);
6232 cache = btrfs_lookup_block_group(info, bytenr);
6235 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6236 BTRFS_BLOCK_GROUP_RAID1 |
6237 BTRFS_BLOCK_GROUP_RAID10))
6242 * If this block group has free space cache written out, we
6243 * need to make sure to load it if we are removing space. This
6244 * is because we need the unpinning stage to actually add the
6245 * space back to the block group, otherwise we will leak space.
6247 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6248 cache_block_group(cache, 1);
6250 byte_in_group = bytenr - cache->key.objectid;
6251 WARN_ON(byte_in_group > cache->key.offset);
6253 spin_lock(&cache->space_info->lock);
6254 spin_lock(&cache->lock);
6256 if (btrfs_test_opt(info, SPACE_CACHE) &&
6257 cache->disk_cache_state < BTRFS_DC_CLEAR)
6258 cache->disk_cache_state = BTRFS_DC_CLEAR;
6260 old_val = btrfs_block_group_used(&cache->item);
6261 num_bytes = min(total, cache->key.offset - byte_in_group);
6263 old_val += num_bytes;
6264 btrfs_set_block_group_used(&cache->item, old_val);
6265 cache->reserved -= num_bytes;
6266 cache->space_info->bytes_reserved -= num_bytes;
6267 cache->space_info->bytes_used += num_bytes;
6268 cache->space_info->disk_used += num_bytes * factor;
6269 spin_unlock(&cache->lock);
6270 spin_unlock(&cache->space_info->lock);
6272 old_val -= num_bytes;
6273 btrfs_set_block_group_used(&cache->item, old_val);
6274 cache->pinned += num_bytes;
6275 cache->space_info->bytes_pinned += num_bytes;
6276 cache->space_info->bytes_used -= num_bytes;
6277 cache->space_info->disk_used -= num_bytes * factor;
6278 spin_unlock(&cache->lock);
6279 spin_unlock(&cache->space_info->lock);
6281 trace_btrfs_space_reservation(info, "pinned",
6282 cache->space_info->flags,
6284 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6286 set_extent_dirty(info->pinned_extents,
6287 bytenr, bytenr + num_bytes - 1,
6288 GFP_NOFS | __GFP_NOFAIL);
6291 spin_lock(&trans->transaction->dirty_bgs_lock);
6292 if (list_empty(&cache->dirty_list)) {
6293 list_add_tail(&cache->dirty_list,
6294 &trans->transaction->dirty_bgs);
6295 trans->transaction->num_dirty_bgs++;
6296 btrfs_get_block_group(cache);
6298 spin_unlock(&trans->transaction->dirty_bgs_lock);
6301 * No longer have used bytes in this block group, queue it for
6302 * deletion. We do this after adding the block group to the
6303 * dirty list to avoid races between cleaner kthread and space
6306 if (!alloc && old_val == 0) {
6307 spin_lock(&info->unused_bgs_lock);
6308 if (list_empty(&cache->bg_list)) {
6309 btrfs_get_block_group(cache);
6310 list_add_tail(&cache->bg_list,
6313 spin_unlock(&info->unused_bgs_lock);
6316 btrfs_put_block_group(cache);
6318 bytenr += num_bytes;
6323 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6325 struct btrfs_block_group_cache *cache;
6328 spin_lock(&fs_info->block_group_cache_lock);
6329 bytenr = fs_info->first_logical_byte;
6330 spin_unlock(&fs_info->block_group_cache_lock);
6332 if (bytenr < (u64)-1)
6335 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6339 bytenr = cache->key.objectid;
6340 btrfs_put_block_group(cache);
6345 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6346 struct btrfs_block_group_cache *cache,
6347 u64 bytenr, u64 num_bytes, int reserved)
6349 spin_lock(&cache->space_info->lock);
6350 spin_lock(&cache->lock);
6351 cache->pinned += num_bytes;
6352 cache->space_info->bytes_pinned += num_bytes;
6354 cache->reserved -= num_bytes;
6355 cache->space_info->bytes_reserved -= num_bytes;
6357 spin_unlock(&cache->lock);
6358 spin_unlock(&cache->space_info->lock);
6360 trace_btrfs_space_reservation(fs_info, "pinned",
6361 cache->space_info->flags, num_bytes, 1);
6362 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6363 set_extent_dirty(fs_info->pinned_extents, bytenr,
6364 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6369 * this function must be called within transaction
6371 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6372 u64 bytenr, u64 num_bytes, int reserved)
6374 struct btrfs_block_group_cache *cache;
6376 cache = btrfs_lookup_block_group(fs_info, bytenr);
6377 BUG_ON(!cache); /* Logic error */
6379 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6381 btrfs_put_block_group(cache);
6386 * this function must be called within transaction
6388 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6389 u64 bytenr, u64 num_bytes)
6391 struct btrfs_block_group_cache *cache;
6394 cache = btrfs_lookup_block_group(fs_info, bytenr);
6399 * pull in the free space cache (if any) so that our pin
6400 * removes the free space from the cache. We have load_only set
6401 * to one because the slow code to read in the free extents does check
6402 * the pinned extents.
6404 cache_block_group(cache, 1);
6406 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6408 /* remove us from the free space cache (if we're there at all) */
6409 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6410 btrfs_put_block_group(cache);
6414 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6415 u64 start, u64 num_bytes)
6418 struct btrfs_block_group_cache *block_group;
6419 struct btrfs_caching_control *caching_ctl;
6421 block_group = btrfs_lookup_block_group(fs_info, start);
6425 cache_block_group(block_group, 0);
6426 caching_ctl = get_caching_control(block_group);
6430 BUG_ON(!block_group_cache_done(block_group));
6431 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6433 mutex_lock(&caching_ctl->mutex);
6435 if (start >= caching_ctl->progress) {
6436 ret = add_excluded_extent(fs_info, start, num_bytes);
6437 } else if (start + num_bytes <= caching_ctl->progress) {
6438 ret = btrfs_remove_free_space(block_group,
6441 num_bytes = caching_ctl->progress - start;
6442 ret = btrfs_remove_free_space(block_group,
6447 num_bytes = (start + num_bytes) -
6448 caching_ctl->progress;
6449 start = caching_ctl->progress;
6450 ret = add_excluded_extent(fs_info, start, num_bytes);
6453 mutex_unlock(&caching_ctl->mutex);
6454 put_caching_control(caching_ctl);
6456 btrfs_put_block_group(block_group);
6460 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6461 struct extent_buffer *eb)
6463 struct btrfs_file_extent_item *item;
6464 struct btrfs_key key;
6468 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6471 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6472 btrfs_item_key_to_cpu(eb, &key, i);
6473 if (key.type != BTRFS_EXTENT_DATA_KEY)
6475 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6476 found_type = btrfs_file_extent_type(eb, item);
6477 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6479 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6481 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6482 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6483 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6490 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6492 atomic_inc(&bg->reservations);
6495 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6498 struct btrfs_block_group_cache *bg;
6500 bg = btrfs_lookup_block_group(fs_info, start);
6502 if (atomic_dec_and_test(&bg->reservations))
6503 wake_up_atomic_t(&bg->reservations);
6504 btrfs_put_block_group(bg);
6507 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6509 struct btrfs_space_info *space_info = bg->space_info;
6513 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6517 * Our block group is read only but before we set it to read only,
6518 * some task might have had allocated an extent from it already, but it
6519 * has not yet created a respective ordered extent (and added it to a
6520 * root's list of ordered extents).
6521 * Therefore wait for any task currently allocating extents, since the
6522 * block group's reservations counter is incremented while a read lock
6523 * on the groups' semaphore is held and decremented after releasing
6524 * the read access on that semaphore and creating the ordered extent.
6526 down_write(&space_info->groups_sem);
6527 up_write(&space_info->groups_sem);
6529 wait_on_atomic_t(&bg->reservations, atomic_t_wait,
6530 TASK_UNINTERRUPTIBLE);
6534 * btrfs_add_reserved_bytes - update the block_group and space info counters
6535 * @cache: The cache we are manipulating
6536 * @ram_bytes: The number of bytes of file content, and will be same to
6537 * @num_bytes except for the compress path.
6538 * @num_bytes: The number of bytes in question
6539 * @delalloc: The blocks are allocated for the delalloc write
6541 * This is called by the allocator when it reserves space. If this is a
6542 * reservation and the block group has become read only we cannot make the
6543 * reservation and return -EAGAIN, otherwise this function always succeeds.
6545 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6546 u64 ram_bytes, u64 num_bytes, int delalloc)
6548 struct btrfs_space_info *space_info = cache->space_info;
6551 spin_lock(&space_info->lock);
6552 spin_lock(&cache->lock);
6556 cache->reserved += num_bytes;
6557 space_info->bytes_reserved += num_bytes;
6559 trace_btrfs_space_reservation(cache->fs_info,
6560 "space_info", space_info->flags,
6562 space_info->bytes_may_use -= ram_bytes;
6564 cache->delalloc_bytes += num_bytes;
6566 spin_unlock(&cache->lock);
6567 spin_unlock(&space_info->lock);
6572 * btrfs_free_reserved_bytes - update the block_group and space info counters
6573 * @cache: The cache we are manipulating
6574 * @num_bytes: The number of bytes in question
6575 * @delalloc: The blocks are allocated for the delalloc write
6577 * This is called by somebody who is freeing space that was never actually used
6578 * on disk. For example if you reserve some space for a new leaf in transaction
6579 * A and before transaction A commits you free that leaf, you call this with
6580 * reserve set to 0 in order to clear the reservation.
6583 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6584 u64 num_bytes, int delalloc)
6586 struct btrfs_space_info *space_info = cache->space_info;
6589 spin_lock(&space_info->lock);
6590 spin_lock(&cache->lock);
6592 space_info->bytes_readonly += num_bytes;
6593 cache->reserved -= num_bytes;
6594 space_info->bytes_reserved -= num_bytes;
6597 cache->delalloc_bytes -= num_bytes;
6598 spin_unlock(&cache->lock);
6599 spin_unlock(&space_info->lock);
6602 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6604 struct btrfs_caching_control *next;
6605 struct btrfs_caching_control *caching_ctl;
6606 struct btrfs_block_group_cache *cache;
6608 down_write(&fs_info->commit_root_sem);
6610 list_for_each_entry_safe(caching_ctl, next,
6611 &fs_info->caching_block_groups, list) {
6612 cache = caching_ctl->block_group;
6613 if (block_group_cache_done(cache)) {
6614 cache->last_byte_to_unpin = (u64)-1;
6615 list_del_init(&caching_ctl->list);
6616 put_caching_control(caching_ctl);
6618 cache->last_byte_to_unpin = caching_ctl->progress;
6622 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6623 fs_info->pinned_extents = &fs_info->freed_extents[1];
6625 fs_info->pinned_extents = &fs_info->freed_extents[0];
6627 up_write(&fs_info->commit_root_sem);
6629 update_global_block_rsv(fs_info);
6633 * Returns the free cluster for the given space info and sets empty_cluster to
6634 * what it should be based on the mount options.
6636 static struct btrfs_free_cluster *
6637 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6638 struct btrfs_space_info *space_info, u64 *empty_cluster)
6640 struct btrfs_free_cluster *ret = NULL;
6643 if (btrfs_mixed_space_info(space_info))
6646 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6647 ret = &fs_info->meta_alloc_cluster;
6648 if (btrfs_test_opt(fs_info, SSD))
6649 *empty_cluster = SZ_2M;
6651 *empty_cluster = SZ_64K;
6652 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6653 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6654 *empty_cluster = SZ_2M;
6655 ret = &fs_info->data_alloc_cluster;
6661 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6663 const bool return_free_space)
6665 struct btrfs_block_group_cache *cache = NULL;
6666 struct btrfs_space_info *space_info;
6667 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6668 struct btrfs_free_cluster *cluster = NULL;
6670 u64 total_unpinned = 0;
6671 u64 empty_cluster = 0;
6674 while (start <= end) {
6677 start >= cache->key.objectid + cache->key.offset) {
6679 btrfs_put_block_group(cache);
6681 cache = btrfs_lookup_block_group(fs_info, start);
6682 BUG_ON(!cache); /* Logic error */
6684 cluster = fetch_cluster_info(fs_info,
6687 empty_cluster <<= 1;
6690 len = cache->key.objectid + cache->key.offset - start;
6691 len = min(len, end + 1 - start);
6693 if (start < cache->last_byte_to_unpin) {
6694 len = min(len, cache->last_byte_to_unpin - start);
6695 if (return_free_space)
6696 btrfs_add_free_space(cache, start, len);
6700 total_unpinned += len;
6701 space_info = cache->space_info;
6704 * If this space cluster has been marked as fragmented and we've
6705 * unpinned enough in this block group to potentially allow a
6706 * cluster to be created inside of it go ahead and clear the
6709 if (cluster && cluster->fragmented &&
6710 total_unpinned > empty_cluster) {
6711 spin_lock(&cluster->lock);
6712 cluster->fragmented = 0;
6713 spin_unlock(&cluster->lock);
6716 spin_lock(&space_info->lock);
6717 spin_lock(&cache->lock);
6718 cache->pinned -= len;
6719 space_info->bytes_pinned -= len;
6721 trace_btrfs_space_reservation(fs_info, "pinned",
6722 space_info->flags, len, 0);
6723 space_info->max_extent_size = 0;
6724 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6726 space_info->bytes_readonly += len;
6729 spin_unlock(&cache->lock);
6730 if (!readonly && return_free_space &&
6731 global_rsv->space_info == space_info) {
6734 spin_lock(&global_rsv->lock);
6735 if (!global_rsv->full) {
6736 to_add = min(len, global_rsv->size -
6737 global_rsv->reserved);
6738 global_rsv->reserved += to_add;
6739 space_info->bytes_may_use += to_add;
6740 if (global_rsv->reserved >= global_rsv->size)
6741 global_rsv->full = 1;
6742 trace_btrfs_space_reservation(fs_info,
6748 spin_unlock(&global_rsv->lock);
6749 /* Add to any tickets we may have */
6751 space_info_add_new_bytes(fs_info, space_info,
6754 spin_unlock(&space_info->lock);
6758 btrfs_put_block_group(cache);
6762 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6764 struct btrfs_fs_info *fs_info = trans->fs_info;
6765 struct btrfs_block_group_cache *block_group, *tmp;
6766 struct list_head *deleted_bgs;
6767 struct extent_io_tree *unpin;
6772 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6773 unpin = &fs_info->freed_extents[1];
6775 unpin = &fs_info->freed_extents[0];
6777 while (!trans->aborted) {
6778 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6779 ret = find_first_extent_bit(unpin, 0, &start, &end,
6780 EXTENT_DIRTY, NULL);
6782 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6786 if (btrfs_test_opt(fs_info, DISCARD))
6787 ret = btrfs_discard_extent(fs_info, start,
6788 end + 1 - start, NULL);
6790 clear_extent_dirty(unpin, start, end);
6791 unpin_extent_range(fs_info, start, end, true);
6792 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6797 * Transaction is finished. We don't need the lock anymore. We
6798 * do need to clean up the block groups in case of a transaction
6801 deleted_bgs = &trans->transaction->deleted_bgs;
6802 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6806 if (!trans->aborted)
6807 ret = btrfs_discard_extent(fs_info,
6808 block_group->key.objectid,
6809 block_group->key.offset,
6812 list_del_init(&block_group->bg_list);
6813 btrfs_put_block_group_trimming(block_group);
6814 btrfs_put_block_group(block_group);
6817 const char *errstr = btrfs_decode_error(ret);
6819 "discard failed while removing blockgroup: errno=%d %s",
6827 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6828 struct btrfs_fs_info *info,
6829 struct btrfs_delayed_ref_node *node, u64 parent,
6830 u64 root_objectid, u64 owner_objectid,
6831 u64 owner_offset, int refs_to_drop,
6832 struct btrfs_delayed_extent_op *extent_op)
6834 struct btrfs_key key;
6835 struct btrfs_path *path;
6836 struct btrfs_root *extent_root = info->extent_root;
6837 struct extent_buffer *leaf;
6838 struct btrfs_extent_item *ei;
6839 struct btrfs_extent_inline_ref *iref;
6842 int extent_slot = 0;
6843 int found_extent = 0;
6847 u64 bytenr = node->bytenr;
6848 u64 num_bytes = node->num_bytes;
6850 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6852 path = btrfs_alloc_path();
6856 path->reada = READA_FORWARD;
6857 path->leave_spinning = 1;
6859 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6860 BUG_ON(!is_data && refs_to_drop != 1);
6863 skinny_metadata = false;
6865 ret = lookup_extent_backref(trans, info, path, &iref,
6866 bytenr, num_bytes, parent,
6867 root_objectid, owner_objectid,
6870 extent_slot = path->slots[0];
6871 while (extent_slot >= 0) {
6872 btrfs_item_key_to_cpu(path->nodes[0], &key,
6874 if (key.objectid != bytenr)
6876 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6877 key.offset == num_bytes) {
6881 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6882 key.offset == owner_objectid) {
6886 if (path->slots[0] - extent_slot > 5)
6890 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6891 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6892 if (found_extent && item_size < sizeof(*ei))
6895 if (!found_extent) {
6897 ret = remove_extent_backref(trans, info, path, NULL,
6899 is_data, &last_ref);
6901 btrfs_abort_transaction(trans, ret);
6904 btrfs_release_path(path);
6905 path->leave_spinning = 1;
6907 key.objectid = bytenr;
6908 key.type = BTRFS_EXTENT_ITEM_KEY;
6909 key.offset = num_bytes;
6911 if (!is_data && skinny_metadata) {
6912 key.type = BTRFS_METADATA_ITEM_KEY;
6913 key.offset = owner_objectid;
6916 ret = btrfs_search_slot(trans, extent_root,
6918 if (ret > 0 && skinny_metadata && path->slots[0]) {
6920 * Couldn't find our skinny metadata item,
6921 * see if we have ye olde extent item.
6924 btrfs_item_key_to_cpu(path->nodes[0], &key,
6926 if (key.objectid == bytenr &&
6927 key.type == BTRFS_EXTENT_ITEM_KEY &&
6928 key.offset == num_bytes)
6932 if (ret > 0 && skinny_metadata) {
6933 skinny_metadata = false;
6934 key.objectid = bytenr;
6935 key.type = BTRFS_EXTENT_ITEM_KEY;
6936 key.offset = num_bytes;
6937 btrfs_release_path(path);
6938 ret = btrfs_search_slot(trans, extent_root,
6944 "umm, got %d back from search, was looking for %llu",
6947 btrfs_print_leaf(path->nodes[0]);
6950 btrfs_abort_transaction(trans, ret);
6953 extent_slot = path->slots[0];
6955 } else if (WARN_ON(ret == -ENOENT)) {
6956 btrfs_print_leaf(path->nodes[0]);
6958 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6959 bytenr, parent, root_objectid, owner_objectid,
6961 btrfs_abort_transaction(trans, ret);
6964 btrfs_abort_transaction(trans, ret);
6968 leaf = path->nodes[0];
6969 item_size = btrfs_item_size_nr(leaf, extent_slot);
6970 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6971 if (item_size < sizeof(*ei)) {
6972 BUG_ON(found_extent || extent_slot != path->slots[0]);
6973 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
6976 btrfs_abort_transaction(trans, ret);
6980 btrfs_release_path(path);
6981 path->leave_spinning = 1;
6983 key.objectid = bytenr;
6984 key.type = BTRFS_EXTENT_ITEM_KEY;
6985 key.offset = num_bytes;
6987 ret = btrfs_search_slot(trans, extent_root, &key, path,
6991 "umm, got %d back from search, was looking for %llu",
6993 btrfs_print_leaf(path->nodes[0]);
6996 btrfs_abort_transaction(trans, ret);
7000 extent_slot = path->slots[0];
7001 leaf = path->nodes[0];
7002 item_size = btrfs_item_size_nr(leaf, extent_slot);
7005 BUG_ON(item_size < sizeof(*ei));
7006 ei = btrfs_item_ptr(leaf, extent_slot,
7007 struct btrfs_extent_item);
7008 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7009 key.type == BTRFS_EXTENT_ITEM_KEY) {
7010 struct btrfs_tree_block_info *bi;
7011 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7012 bi = (struct btrfs_tree_block_info *)(ei + 1);
7013 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7016 refs = btrfs_extent_refs(leaf, ei);
7017 if (refs < refs_to_drop) {
7019 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7020 refs_to_drop, refs, bytenr);
7022 btrfs_abort_transaction(trans, ret);
7025 refs -= refs_to_drop;
7029 __run_delayed_extent_op(extent_op, leaf, ei);
7031 * In the case of inline back ref, reference count will
7032 * be updated by remove_extent_backref
7035 BUG_ON(!found_extent);
7037 btrfs_set_extent_refs(leaf, ei, refs);
7038 btrfs_mark_buffer_dirty(leaf);
7041 ret = remove_extent_backref(trans, info, path,
7043 is_data, &last_ref);
7045 btrfs_abort_transaction(trans, ret);
7051 BUG_ON(is_data && refs_to_drop !=
7052 extent_data_ref_count(path, iref));
7054 BUG_ON(path->slots[0] != extent_slot);
7056 BUG_ON(path->slots[0] != extent_slot + 1);
7057 path->slots[0] = extent_slot;
7063 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7066 btrfs_abort_transaction(trans, ret);
7069 btrfs_release_path(path);
7072 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7074 btrfs_abort_transaction(trans, ret);
7079 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7081 btrfs_abort_transaction(trans, ret);
7085 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7087 btrfs_abort_transaction(trans, ret);
7091 btrfs_release_path(path);
7094 btrfs_free_path(path);
7099 * when we free an block, it is possible (and likely) that we free the last
7100 * delayed ref for that extent as well. This searches the delayed ref tree for
7101 * a given extent, and if there are no other delayed refs to be processed, it
7102 * removes it from the tree.
7104 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7107 struct btrfs_delayed_ref_head *head;
7108 struct btrfs_delayed_ref_root *delayed_refs;
7111 delayed_refs = &trans->transaction->delayed_refs;
7112 spin_lock(&delayed_refs->lock);
7113 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7115 goto out_delayed_unlock;
7117 spin_lock(&head->lock);
7118 if (!RB_EMPTY_ROOT(&head->ref_tree))
7121 if (head->extent_op) {
7122 if (!head->must_insert_reserved)
7124 btrfs_free_delayed_extent_op(head->extent_op);
7125 head->extent_op = NULL;
7129 * waiting for the lock here would deadlock. If someone else has it
7130 * locked they are already in the process of dropping it anyway
7132 if (!mutex_trylock(&head->mutex))
7136 * at this point we have a head with no other entries. Go
7137 * ahead and process it.
7139 rb_erase(&head->href_node, &delayed_refs->href_root);
7140 RB_CLEAR_NODE(&head->href_node);
7141 atomic_dec(&delayed_refs->num_entries);
7144 * we don't take a ref on the node because we're removing it from the
7145 * tree, so we just steal the ref the tree was holding.
7147 delayed_refs->num_heads--;
7148 if (head->processing == 0)
7149 delayed_refs->num_heads_ready--;
7150 head->processing = 0;
7151 spin_unlock(&head->lock);
7152 spin_unlock(&delayed_refs->lock);
7154 BUG_ON(head->extent_op);
7155 if (head->must_insert_reserved)
7158 mutex_unlock(&head->mutex);
7159 btrfs_put_delayed_ref_head(head);
7162 spin_unlock(&head->lock);
7165 spin_unlock(&delayed_refs->lock);
7169 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7170 struct btrfs_root *root,
7171 struct extent_buffer *buf,
7172 u64 parent, int last_ref)
7174 struct btrfs_fs_info *fs_info = root->fs_info;
7178 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7179 int old_ref_mod, new_ref_mod;
7181 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7182 root->root_key.objectid,
7183 btrfs_header_level(buf), 0,
7184 BTRFS_DROP_DELAYED_REF);
7185 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7187 root->root_key.objectid,
7188 btrfs_header_level(buf),
7189 BTRFS_DROP_DELAYED_REF, NULL,
7190 &old_ref_mod, &new_ref_mod);
7191 BUG_ON(ret); /* -ENOMEM */
7192 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7195 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7196 struct btrfs_block_group_cache *cache;
7198 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7199 ret = check_ref_cleanup(trans, buf->start);
7205 cache = btrfs_lookup_block_group(fs_info, buf->start);
7207 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7208 pin_down_extent(fs_info, cache, buf->start,
7210 btrfs_put_block_group(cache);
7214 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7216 btrfs_add_free_space(cache, buf->start, buf->len);
7217 btrfs_free_reserved_bytes(cache, buf->len, 0);
7218 btrfs_put_block_group(cache);
7219 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7223 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7224 root->root_key.objectid);
7228 * Deleting the buffer, clear the corrupt flag since it doesn't
7231 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7235 /* Can return -ENOMEM */
7236 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7237 struct btrfs_root *root,
7238 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7239 u64 owner, u64 offset)
7241 struct btrfs_fs_info *fs_info = root->fs_info;
7242 int old_ref_mod, new_ref_mod;
7245 if (btrfs_is_testing(fs_info))
7248 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7249 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7250 root_objectid, owner, offset,
7251 BTRFS_DROP_DELAYED_REF);
7254 * tree log blocks never actually go into the extent allocation
7255 * tree, just update pinning info and exit early.
7257 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7258 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7259 /* unlocks the pinned mutex */
7260 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7261 old_ref_mod = new_ref_mod = 0;
7263 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7264 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7266 root_objectid, (int)owner,
7267 BTRFS_DROP_DELAYED_REF, NULL,
7268 &old_ref_mod, &new_ref_mod);
7270 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7272 root_objectid, owner, offset,
7273 0, BTRFS_DROP_DELAYED_REF,
7274 &old_ref_mod, &new_ref_mod);
7277 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7278 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7284 * when we wait for progress in the block group caching, its because
7285 * our allocation attempt failed at least once. So, we must sleep
7286 * and let some progress happen before we try again.
7288 * This function will sleep at least once waiting for new free space to
7289 * show up, and then it will check the block group free space numbers
7290 * for our min num_bytes. Another option is to have it go ahead
7291 * and look in the rbtree for a free extent of a given size, but this
7294 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7295 * any of the information in this block group.
7297 static noinline void
7298 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7301 struct btrfs_caching_control *caching_ctl;
7303 caching_ctl = get_caching_control(cache);
7307 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7308 (cache->free_space_ctl->free_space >= num_bytes));
7310 put_caching_control(caching_ctl);
7314 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7316 struct btrfs_caching_control *caching_ctl;
7319 caching_ctl = get_caching_control(cache);
7321 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7323 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7324 if (cache->cached == BTRFS_CACHE_ERROR)
7326 put_caching_control(caching_ctl);
7330 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7331 [BTRFS_RAID_RAID10] = "raid10",
7332 [BTRFS_RAID_RAID1] = "raid1",
7333 [BTRFS_RAID_DUP] = "dup",
7334 [BTRFS_RAID_RAID0] = "raid0",
7335 [BTRFS_RAID_SINGLE] = "single",
7336 [BTRFS_RAID_RAID5] = "raid5",
7337 [BTRFS_RAID_RAID6] = "raid6",
7340 static const char *get_raid_name(enum btrfs_raid_types type)
7342 if (type >= BTRFS_NR_RAID_TYPES)
7345 return btrfs_raid_type_names[type];
7348 enum btrfs_loop_type {
7349 LOOP_CACHING_NOWAIT = 0,
7350 LOOP_CACHING_WAIT = 1,
7351 LOOP_ALLOC_CHUNK = 2,
7352 LOOP_NO_EMPTY_SIZE = 3,
7356 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7360 down_read(&cache->data_rwsem);
7364 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7367 btrfs_get_block_group(cache);
7369 down_read(&cache->data_rwsem);
7372 static struct btrfs_block_group_cache *
7373 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7374 struct btrfs_free_cluster *cluster,
7377 struct btrfs_block_group_cache *used_bg = NULL;
7379 spin_lock(&cluster->refill_lock);
7381 used_bg = cluster->block_group;
7385 if (used_bg == block_group)
7388 btrfs_get_block_group(used_bg);
7393 if (down_read_trylock(&used_bg->data_rwsem))
7396 spin_unlock(&cluster->refill_lock);
7398 /* We should only have one-level nested. */
7399 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7401 spin_lock(&cluster->refill_lock);
7402 if (used_bg == cluster->block_group)
7405 up_read(&used_bg->data_rwsem);
7406 btrfs_put_block_group(used_bg);
7411 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7415 up_read(&cache->data_rwsem);
7416 btrfs_put_block_group(cache);
7420 * walks the btree of allocated extents and find a hole of a given size.
7421 * The key ins is changed to record the hole:
7422 * ins->objectid == start position
7423 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7424 * ins->offset == the size of the hole.
7425 * Any available blocks before search_start are skipped.
7427 * If there is no suitable free space, we will record the max size of
7428 * the free space extent currently.
7430 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7431 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7432 u64 hint_byte, struct btrfs_key *ins,
7433 u64 flags, int delalloc)
7436 struct btrfs_root *root = fs_info->extent_root;
7437 struct btrfs_free_cluster *last_ptr = NULL;
7438 struct btrfs_block_group_cache *block_group = NULL;
7439 u64 search_start = 0;
7440 u64 max_extent_size = 0;
7441 u64 empty_cluster = 0;
7442 struct btrfs_space_info *space_info;
7444 int index = btrfs_bg_flags_to_raid_index(flags);
7445 bool failed_cluster_refill = false;
7446 bool failed_alloc = false;
7447 bool use_cluster = true;
7448 bool have_caching_bg = false;
7449 bool orig_have_caching_bg = false;
7450 bool full_search = false;
7452 WARN_ON(num_bytes < fs_info->sectorsize);
7453 ins->type = BTRFS_EXTENT_ITEM_KEY;
7457 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7459 space_info = __find_space_info(fs_info, flags);
7461 btrfs_err(fs_info, "No space info for %llu", flags);
7466 * If our free space is heavily fragmented we may not be able to make
7467 * big contiguous allocations, so instead of doing the expensive search
7468 * for free space, simply return ENOSPC with our max_extent_size so we
7469 * can go ahead and search for a more manageable chunk.
7471 * If our max_extent_size is large enough for our allocation simply
7472 * disable clustering since we will likely not be able to find enough
7473 * space to create a cluster and induce latency trying.
7475 if (unlikely(space_info->max_extent_size)) {
7476 spin_lock(&space_info->lock);
7477 if (space_info->max_extent_size &&
7478 num_bytes > space_info->max_extent_size) {
7479 ins->offset = space_info->max_extent_size;
7480 spin_unlock(&space_info->lock);
7482 } else if (space_info->max_extent_size) {
7483 use_cluster = false;
7485 spin_unlock(&space_info->lock);
7488 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7490 spin_lock(&last_ptr->lock);
7491 if (last_ptr->block_group)
7492 hint_byte = last_ptr->window_start;
7493 if (last_ptr->fragmented) {
7495 * We still set window_start so we can keep track of the
7496 * last place we found an allocation to try and save
7499 hint_byte = last_ptr->window_start;
7500 use_cluster = false;
7502 spin_unlock(&last_ptr->lock);
7505 search_start = max(search_start, first_logical_byte(fs_info, 0));
7506 search_start = max(search_start, hint_byte);
7507 if (search_start == hint_byte) {
7508 block_group = btrfs_lookup_block_group(fs_info, search_start);
7510 * we don't want to use the block group if it doesn't match our
7511 * allocation bits, or if its not cached.
7513 * However if we are re-searching with an ideal block group
7514 * picked out then we don't care that the block group is cached.
7516 if (block_group && block_group_bits(block_group, flags) &&
7517 block_group->cached != BTRFS_CACHE_NO) {
7518 down_read(&space_info->groups_sem);
7519 if (list_empty(&block_group->list) ||
7522 * someone is removing this block group,
7523 * we can't jump into the have_block_group
7524 * target because our list pointers are not
7527 btrfs_put_block_group(block_group);
7528 up_read(&space_info->groups_sem);
7530 index = btrfs_bg_flags_to_raid_index(
7531 block_group->flags);
7532 btrfs_lock_block_group(block_group, delalloc);
7533 goto have_block_group;
7535 } else if (block_group) {
7536 btrfs_put_block_group(block_group);
7540 have_caching_bg = false;
7541 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7543 down_read(&space_info->groups_sem);
7544 list_for_each_entry(block_group, &space_info->block_groups[index],
7549 /* If the block group is read-only, we can skip it entirely. */
7550 if (unlikely(block_group->ro))
7553 btrfs_grab_block_group(block_group, delalloc);
7554 search_start = block_group->key.objectid;
7557 * this can happen if we end up cycling through all the
7558 * raid types, but we want to make sure we only allocate
7559 * for the proper type.
7561 if (!block_group_bits(block_group, flags)) {
7562 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7563 BTRFS_BLOCK_GROUP_RAID1 |
7564 BTRFS_BLOCK_GROUP_RAID5 |
7565 BTRFS_BLOCK_GROUP_RAID6 |
7566 BTRFS_BLOCK_GROUP_RAID10;
7569 * if they asked for extra copies and this block group
7570 * doesn't provide them, bail. This does allow us to
7571 * fill raid0 from raid1.
7573 if ((flags & extra) && !(block_group->flags & extra))
7578 cached = block_group_cache_done(block_group);
7579 if (unlikely(!cached)) {
7580 have_caching_bg = true;
7581 ret = cache_block_group(block_group, 0);
7586 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7590 * Ok we want to try and use the cluster allocator, so
7593 if (last_ptr && use_cluster) {
7594 struct btrfs_block_group_cache *used_block_group;
7595 unsigned long aligned_cluster;
7597 * the refill lock keeps out other
7598 * people trying to start a new cluster
7600 used_block_group = btrfs_lock_cluster(block_group,
7603 if (!used_block_group)
7604 goto refill_cluster;
7606 if (used_block_group != block_group &&
7607 (used_block_group->ro ||
7608 !block_group_bits(used_block_group, flags)))
7609 goto release_cluster;
7611 offset = btrfs_alloc_from_cluster(used_block_group,
7614 used_block_group->key.objectid,
7617 /* we have a block, we're done */
7618 spin_unlock(&last_ptr->refill_lock);
7619 trace_btrfs_reserve_extent_cluster(fs_info,
7621 search_start, num_bytes);
7622 if (used_block_group != block_group) {
7623 btrfs_release_block_group(block_group,
7625 block_group = used_block_group;
7630 WARN_ON(last_ptr->block_group != used_block_group);
7632 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7633 * set up a new clusters, so lets just skip it
7634 * and let the allocator find whatever block
7635 * it can find. If we reach this point, we
7636 * will have tried the cluster allocator
7637 * plenty of times and not have found
7638 * anything, so we are likely way too
7639 * fragmented for the clustering stuff to find
7642 * However, if the cluster is taken from the
7643 * current block group, release the cluster
7644 * first, so that we stand a better chance of
7645 * succeeding in the unclustered
7647 if (loop >= LOOP_NO_EMPTY_SIZE &&
7648 used_block_group != block_group) {
7649 spin_unlock(&last_ptr->refill_lock);
7650 btrfs_release_block_group(used_block_group,
7652 goto unclustered_alloc;
7656 * this cluster didn't work out, free it and
7659 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7661 if (used_block_group != block_group)
7662 btrfs_release_block_group(used_block_group,
7665 if (loop >= LOOP_NO_EMPTY_SIZE) {
7666 spin_unlock(&last_ptr->refill_lock);
7667 goto unclustered_alloc;
7670 aligned_cluster = max_t(unsigned long,
7671 empty_cluster + empty_size,
7672 block_group->full_stripe_len);
7674 /* allocate a cluster in this block group */
7675 ret = btrfs_find_space_cluster(fs_info, block_group,
7676 last_ptr, search_start,
7681 * now pull our allocation out of this
7684 offset = btrfs_alloc_from_cluster(block_group,
7690 /* we found one, proceed */
7691 spin_unlock(&last_ptr->refill_lock);
7692 trace_btrfs_reserve_extent_cluster(fs_info,
7693 block_group, search_start,
7697 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7698 && !failed_cluster_refill) {
7699 spin_unlock(&last_ptr->refill_lock);
7701 failed_cluster_refill = true;
7702 wait_block_group_cache_progress(block_group,
7703 num_bytes + empty_cluster + empty_size);
7704 goto have_block_group;
7708 * at this point we either didn't find a cluster
7709 * or we weren't able to allocate a block from our
7710 * cluster. Free the cluster we've been trying
7711 * to use, and go to the next block group
7713 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7714 spin_unlock(&last_ptr->refill_lock);
7720 * We are doing an unclustered alloc, set the fragmented flag so
7721 * we don't bother trying to setup a cluster again until we get
7724 if (unlikely(last_ptr)) {
7725 spin_lock(&last_ptr->lock);
7726 last_ptr->fragmented = 1;
7727 spin_unlock(&last_ptr->lock);
7730 struct btrfs_free_space_ctl *ctl =
7731 block_group->free_space_ctl;
7733 spin_lock(&ctl->tree_lock);
7734 if (ctl->free_space <
7735 num_bytes + empty_cluster + empty_size) {
7736 if (ctl->free_space > max_extent_size)
7737 max_extent_size = ctl->free_space;
7738 spin_unlock(&ctl->tree_lock);
7741 spin_unlock(&ctl->tree_lock);
7744 offset = btrfs_find_space_for_alloc(block_group, search_start,
7745 num_bytes, empty_size,
7748 * If we didn't find a chunk, and we haven't failed on this
7749 * block group before, and this block group is in the middle of
7750 * caching and we are ok with waiting, then go ahead and wait
7751 * for progress to be made, and set failed_alloc to true.
7753 * If failed_alloc is true then we've already waited on this
7754 * block group once and should move on to the next block group.
7756 if (!offset && !failed_alloc && !cached &&
7757 loop > LOOP_CACHING_NOWAIT) {
7758 wait_block_group_cache_progress(block_group,
7759 num_bytes + empty_size);
7760 failed_alloc = true;
7761 goto have_block_group;
7762 } else if (!offset) {
7766 search_start = ALIGN(offset, fs_info->stripesize);
7768 /* move on to the next group */
7769 if (search_start + num_bytes >
7770 block_group->key.objectid + block_group->key.offset) {
7771 btrfs_add_free_space(block_group, offset, num_bytes);
7775 if (offset < search_start)
7776 btrfs_add_free_space(block_group, offset,
7777 search_start - offset);
7778 BUG_ON(offset > search_start);
7780 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7781 num_bytes, delalloc);
7782 if (ret == -EAGAIN) {
7783 btrfs_add_free_space(block_group, offset, num_bytes);
7786 btrfs_inc_block_group_reservations(block_group);
7788 /* we are all good, lets return */
7789 ins->objectid = search_start;
7790 ins->offset = num_bytes;
7792 trace_btrfs_reserve_extent(fs_info, block_group,
7793 search_start, num_bytes);
7794 btrfs_release_block_group(block_group, delalloc);
7797 failed_cluster_refill = false;
7798 failed_alloc = false;
7799 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7801 btrfs_release_block_group(block_group, delalloc);
7804 up_read(&space_info->groups_sem);
7806 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7807 && !orig_have_caching_bg)
7808 orig_have_caching_bg = true;
7810 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7813 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7817 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7818 * caching kthreads as we move along
7819 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7820 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7821 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7824 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7826 if (loop == LOOP_CACHING_NOWAIT) {
7828 * We want to skip the LOOP_CACHING_WAIT step if we
7829 * don't have any uncached bgs and we've already done a
7830 * full search through.
7832 if (orig_have_caching_bg || !full_search)
7833 loop = LOOP_CACHING_WAIT;
7835 loop = LOOP_ALLOC_CHUNK;
7840 if (loop == LOOP_ALLOC_CHUNK) {
7841 struct btrfs_trans_handle *trans;
7844 trans = current->journal_info;
7848 trans = btrfs_join_transaction(root);
7850 if (IS_ERR(trans)) {
7851 ret = PTR_ERR(trans);
7855 ret = do_chunk_alloc(trans, fs_info, flags,
7859 * If we can't allocate a new chunk we've already looped
7860 * through at least once, move on to the NO_EMPTY_SIZE
7864 loop = LOOP_NO_EMPTY_SIZE;
7867 * Do not bail out on ENOSPC since we
7868 * can do more things.
7870 if (ret < 0 && ret != -ENOSPC)
7871 btrfs_abort_transaction(trans, ret);
7875 btrfs_end_transaction(trans);
7880 if (loop == LOOP_NO_EMPTY_SIZE) {
7882 * Don't loop again if we already have no empty_size and
7885 if (empty_size == 0 &&
7886 empty_cluster == 0) {
7895 } else if (!ins->objectid) {
7897 } else if (ins->objectid) {
7898 if (!use_cluster && last_ptr) {
7899 spin_lock(&last_ptr->lock);
7900 last_ptr->window_start = ins->objectid;
7901 spin_unlock(&last_ptr->lock);
7906 if (ret == -ENOSPC) {
7907 spin_lock(&space_info->lock);
7908 space_info->max_extent_size = max_extent_size;
7909 spin_unlock(&space_info->lock);
7910 ins->offset = max_extent_size;
7915 static void dump_space_info(struct btrfs_fs_info *fs_info,
7916 struct btrfs_space_info *info, u64 bytes,
7917 int dump_block_groups)
7919 struct btrfs_block_group_cache *cache;
7922 spin_lock(&info->lock);
7923 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7925 info->total_bytes - btrfs_space_info_used(info, true),
7926 info->full ? "" : "not ");
7928 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7929 info->total_bytes, info->bytes_used, info->bytes_pinned,
7930 info->bytes_reserved, info->bytes_may_use,
7931 info->bytes_readonly);
7932 spin_unlock(&info->lock);
7934 if (!dump_block_groups)
7937 down_read(&info->groups_sem);
7939 list_for_each_entry(cache, &info->block_groups[index], list) {
7940 spin_lock(&cache->lock);
7942 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7943 cache->key.objectid, cache->key.offset,
7944 btrfs_block_group_used(&cache->item), cache->pinned,
7945 cache->reserved, cache->ro ? "[readonly]" : "");
7946 btrfs_dump_free_space(cache, bytes);
7947 spin_unlock(&cache->lock);
7949 if (++index < BTRFS_NR_RAID_TYPES)
7951 up_read(&info->groups_sem);
7955 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7956 * hole that is at least as big as @num_bytes.
7958 * @root - The root that will contain this extent
7960 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7961 * is used for accounting purposes. This value differs
7962 * from @num_bytes only in the case of compressed extents.
7964 * @num_bytes - Number of bytes to allocate on-disk.
7966 * @min_alloc_size - Indicates the minimum amount of space that the
7967 * allocator should try to satisfy. In some cases
7968 * @num_bytes may be larger than what is required and if
7969 * the filesystem is fragmented then allocation fails.
7970 * However, the presence of @min_alloc_size gives a
7971 * chance to try and satisfy the smaller allocation.
7973 * @empty_size - A hint that you plan on doing more COW. This is the
7974 * size in bytes the allocator should try to find free
7975 * next to the block it returns. This is just a hint and
7976 * may be ignored by the allocator.
7978 * @hint_byte - Hint to the allocator to start searching above the byte
7979 * address passed. It might be ignored.
7981 * @ins - This key is modified to record the found hole. It will
7982 * have the following values:
7983 * ins->objectid == start position
7984 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7985 * ins->offset == the size of the hole.
7987 * @is_data - Boolean flag indicating whether an extent is
7988 * allocated for data (true) or metadata (false)
7990 * @delalloc - Boolean flag indicating whether this allocation is for
7991 * delalloc or not. If 'true' data_rwsem of block groups
7992 * is going to be acquired.
7995 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7996 * case -ENOSPC is returned then @ins->offset will contain the size of the
7997 * largest available hole the allocator managed to find.
7999 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8000 u64 num_bytes, u64 min_alloc_size,
8001 u64 empty_size, u64 hint_byte,
8002 struct btrfs_key *ins, int is_data, int delalloc)
8004 struct btrfs_fs_info *fs_info = root->fs_info;
8005 bool final_tried = num_bytes == min_alloc_size;
8009 flags = get_alloc_profile_by_root(root, is_data);
8011 WARN_ON(num_bytes < fs_info->sectorsize);
8012 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8013 hint_byte, ins, flags, delalloc);
8014 if (!ret && !is_data) {
8015 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8016 } else if (ret == -ENOSPC) {
8017 if (!final_tried && ins->offset) {
8018 num_bytes = min(num_bytes >> 1, ins->offset);
8019 num_bytes = round_down(num_bytes,
8020 fs_info->sectorsize);
8021 num_bytes = max(num_bytes, min_alloc_size);
8022 ram_bytes = num_bytes;
8023 if (num_bytes == min_alloc_size)
8026 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8027 struct btrfs_space_info *sinfo;
8029 sinfo = __find_space_info(fs_info, flags);
8031 "allocation failed flags %llu, wanted %llu",
8034 dump_space_info(fs_info, sinfo, num_bytes, 1);
8041 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8043 int pin, int delalloc)
8045 struct btrfs_block_group_cache *cache;
8048 cache = btrfs_lookup_block_group(fs_info, start);
8050 btrfs_err(fs_info, "Unable to find block group for %llu",
8056 pin_down_extent(fs_info, cache, start, len, 1);
8058 if (btrfs_test_opt(fs_info, DISCARD))
8059 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8060 btrfs_add_free_space(cache, start, len);
8061 btrfs_free_reserved_bytes(cache, len, delalloc);
8062 trace_btrfs_reserved_extent_free(fs_info, start, len);
8065 btrfs_put_block_group(cache);
8069 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8070 u64 start, u64 len, int delalloc)
8072 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8075 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8078 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8081 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8082 struct btrfs_fs_info *fs_info,
8083 u64 parent, u64 root_objectid,
8084 u64 flags, u64 owner, u64 offset,
8085 struct btrfs_key *ins, int ref_mod)
8088 struct btrfs_extent_item *extent_item;
8089 struct btrfs_extent_inline_ref *iref;
8090 struct btrfs_path *path;
8091 struct extent_buffer *leaf;
8096 type = BTRFS_SHARED_DATA_REF_KEY;
8098 type = BTRFS_EXTENT_DATA_REF_KEY;
8100 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8102 path = btrfs_alloc_path();
8106 path->leave_spinning = 1;
8107 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8110 btrfs_free_path(path);
8114 leaf = path->nodes[0];
8115 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8116 struct btrfs_extent_item);
8117 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8118 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8119 btrfs_set_extent_flags(leaf, extent_item,
8120 flags | BTRFS_EXTENT_FLAG_DATA);
8122 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8123 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8125 struct btrfs_shared_data_ref *ref;
8126 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8127 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8128 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8130 struct btrfs_extent_data_ref *ref;
8131 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8132 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8133 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8134 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8135 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8138 btrfs_mark_buffer_dirty(path->nodes[0]);
8139 btrfs_free_path(path);
8141 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8146 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8147 if (ret) { /* -ENOENT, logic error */
8148 btrfs_err(fs_info, "update block group failed for %llu %llu",
8149 ins->objectid, ins->offset);
8152 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8156 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8157 struct btrfs_fs_info *fs_info,
8158 u64 parent, u64 root_objectid,
8159 u64 flags, struct btrfs_disk_key *key,
8160 int level, struct btrfs_key *ins)
8163 struct btrfs_extent_item *extent_item;
8164 struct btrfs_tree_block_info *block_info;
8165 struct btrfs_extent_inline_ref *iref;
8166 struct btrfs_path *path;
8167 struct extent_buffer *leaf;
8168 u32 size = sizeof(*extent_item) + sizeof(*iref);
8169 u64 num_bytes = ins->offset;
8170 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8172 if (!skinny_metadata)
8173 size += sizeof(*block_info);
8175 path = btrfs_alloc_path();
8177 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8182 path->leave_spinning = 1;
8183 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8186 btrfs_free_path(path);
8187 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8192 leaf = path->nodes[0];
8193 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8194 struct btrfs_extent_item);
8195 btrfs_set_extent_refs(leaf, extent_item, 1);
8196 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8197 btrfs_set_extent_flags(leaf, extent_item,
8198 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8200 if (skinny_metadata) {
8201 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8202 num_bytes = fs_info->nodesize;
8204 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8205 btrfs_set_tree_block_key(leaf, block_info, key);
8206 btrfs_set_tree_block_level(leaf, block_info, level);
8207 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8211 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8212 btrfs_set_extent_inline_ref_type(leaf, iref,
8213 BTRFS_SHARED_BLOCK_REF_KEY);
8214 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8216 btrfs_set_extent_inline_ref_type(leaf, iref,
8217 BTRFS_TREE_BLOCK_REF_KEY);
8218 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8221 btrfs_mark_buffer_dirty(leaf);
8222 btrfs_free_path(path);
8224 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8229 ret = update_block_group(trans, fs_info, ins->objectid,
8230 fs_info->nodesize, 1);
8231 if (ret) { /* -ENOENT, logic error */
8232 btrfs_err(fs_info, "update block group failed for %llu %llu",
8233 ins->objectid, ins->offset);
8237 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8242 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8243 struct btrfs_root *root, u64 owner,
8244 u64 offset, u64 ram_bytes,
8245 struct btrfs_key *ins)
8247 struct btrfs_fs_info *fs_info = root->fs_info;
8250 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8252 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8253 root->root_key.objectid, owner, offset,
8254 BTRFS_ADD_DELAYED_EXTENT);
8256 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8258 root->root_key.objectid, owner,
8260 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8265 * this is used by the tree logging recovery code. It records that
8266 * an extent has been allocated and makes sure to clear the free
8267 * space cache bits as well
8269 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8270 struct btrfs_fs_info *fs_info,
8271 u64 root_objectid, u64 owner, u64 offset,
8272 struct btrfs_key *ins)
8275 struct btrfs_block_group_cache *block_group;
8276 struct btrfs_space_info *space_info;
8279 * Mixed block groups will exclude before processing the log so we only
8280 * need to do the exclude dance if this fs isn't mixed.
8282 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8283 ret = __exclude_logged_extent(fs_info, ins->objectid,
8289 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8293 space_info = block_group->space_info;
8294 spin_lock(&space_info->lock);
8295 spin_lock(&block_group->lock);
8296 space_info->bytes_reserved += ins->offset;
8297 block_group->reserved += ins->offset;
8298 spin_unlock(&block_group->lock);
8299 spin_unlock(&space_info->lock);
8301 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8302 0, owner, offset, ins, 1);
8303 btrfs_put_block_group(block_group);
8307 static struct extent_buffer *
8308 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8309 u64 bytenr, int level)
8311 struct btrfs_fs_info *fs_info = root->fs_info;
8312 struct extent_buffer *buf;
8314 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8318 btrfs_set_header_generation(buf, trans->transid);
8319 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8320 btrfs_tree_lock(buf);
8321 clean_tree_block(fs_info, buf);
8322 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8324 btrfs_set_lock_blocking(buf);
8325 set_extent_buffer_uptodate(buf);
8327 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8328 buf->log_index = root->log_transid % 2;
8330 * we allow two log transactions at a time, use different
8331 * EXENT bit to differentiate dirty pages.
8333 if (buf->log_index == 0)
8334 set_extent_dirty(&root->dirty_log_pages, buf->start,
8335 buf->start + buf->len - 1, GFP_NOFS);
8337 set_extent_new(&root->dirty_log_pages, buf->start,
8338 buf->start + buf->len - 1);
8340 buf->log_index = -1;
8341 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8342 buf->start + buf->len - 1, GFP_NOFS);
8344 trans->dirty = true;
8345 /* this returns a buffer locked for blocking */
8349 static struct btrfs_block_rsv *
8350 use_block_rsv(struct btrfs_trans_handle *trans,
8351 struct btrfs_root *root, u32 blocksize)
8353 struct btrfs_fs_info *fs_info = root->fs_info;
8354 struct btrfs_block_rsv *block_rsv;
8355 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8357 bool global_updated = false;
8359 block_rsv = get_block_rsv(trans, root);
8361 if (unlikely(block_rsv->size == 0))
8364 ret = block_rsv_use_bytes(block_rsv, blocksize);
8368 if (block_rsv->failfast)
8369 return ERR_PTR(ret);
8371 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8372 global_updated = true;
8373 update_global_block_rsv(fs_info);
8377 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8378 static DEFINE_RATELIMIT_STATE(_rs,
8379 DEFAULT_RATELIMIT_INTERVAL * 10,
8380 /*DEFAULT_RATELIMIT_BURST*/ 1);
8381 if (__ratelimit(&_rs))
8383 "BTRFS: block rsv returned %d\n", ret);
8386 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8387 BTRFS_RESERVE_NO_FLUSH);
8391 * If we couldn't reserve metadata bytes try and use some from
8392 * the global reserve if its space type is the same as the global
8395 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8396 block_rsv->space_info == global_rsv->space_info) {
8397 ret = block_rsv_use_bytes(global_rsv, blocksize);
8401 return ERR_PTR(ret);
8404 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8405 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8407 block_rsv_add_bytes(block_rsv, blocksize, 0);
8408 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8412 * finds a free extent and does all the dirty work required for allocation
8413 * returns the tree buffer or an ERR_PTR on error.
8415 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8416 struct btrfs_root *root,
8417 u64 parent, u64 root_objectid,
8418 const struct btrfs_disk_key *key,
8419 int level, u64 hint,
8422 struct btrfs_fs_info *fs_info = root->fs_info;
8423 struct btrfs_key ins;
8424 struct btrfs_block_rsv *block_rsv;
8425 struct extent_buffer *buf;
8426 struct btrfs_delayed_extent_op *extent_op;
8429 u32 blocksize = fs_info->nodesize;
8430 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8432 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8433 if (btrfs_is_testing(fs_info)) {
8434 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8437 root->alloc_bytenr += blocksize;
8442 block_rsv = use_block_rsv(trans, root, blocksize);
8443 if (IS_ERR(block_rsv))
8444 return ERR_CAST(block_rsv);
8446 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8447 empty_size, hint, &ins, 0, 0);
8451 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8454 goto out_free_reserved;
8457 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8459 parent = ins.objectid;
8460 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8464 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8465 extent_op = btrfs_alloc_delayed_extent_op();
8471 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8473 memset(&extent_op->key, 0, sizeof(extent_op->key));
8474 extent_op->flags_to_set = flags;
8475 extent_op->update_key = skinny_metadata ? false : true;
8476 extent_op->update_flags = true;
8477 extent_op->is_data = false;
8478 extent_op->level = level;
8480 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8481 root_objectid, level, 0,
8482 BTRFS_ADD_DELAYED_EXTENT);
8483 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8485 root_objectid, level,
8486 BTRFS_ADD_DELAYED_EXTENT,
8487 extent_op, NULL, NULL);
8489 goto out_free_delayed;
8494 btrfs_free_delayed_extent_op(extent_op);
8496 free_extent_buffer(buf);
8498 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8500 unuse_block_rsv(fs_info, block_rsv, blocksize);
8501 return ERR_PTR(ret);
8504 struct walk_control {
8505 u64 refs[BTRFS_MAX_LEVEL];
8506 u64 flags[BTRFS_MAX_LEVEL];
8507 struct btrfs_key update_progress;
8518 #define DROP_REFERENCE 1
8519 #define UPDATE_BACKREF 2
8521 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8522 struct btrfs_root *root,
8523 struct walk_control *wc,
8524 struct btrfs_path *path)
8526 struct btrfs_fs_info *fs_info = root->fs_info;
8532 struct btrfs_key key;
8533 struct extent_buffer *eb;
8538 if (path->slots[wc->level] < wc->reada_slot) {
8539 wc->reada_count = wc->reada_count * 2 / 3;
8540 wc->reada_count = max(wc->reada_count, 2);
8542 wc->reada_count = wc->reada_count * 3 / 2;
8543 wc->reada_count = min_t(int, wc->reada_count,
8544 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8547 eb = path->nodes[wc->level];
8548 nritems = btrfs_header_nritems(eb);
8550 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8551 if (nread >= wc->reada_count)
8555 bytenr = btrfs_node_blockptr(eb, slot);
8556 generation = btrfs_node_ptr_generation(eb, slot);
8558 if (slot == path->slots[wc->level])
8561 if (wc->stage == UPDATE_BACKREF &&
8562 generation <= root->root_key.offset)
8565 /* We don't lock the tree block, it's OK to be racy here */
8566 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8567 wc->level - 1, 1, &refs,
8569 /* We don't care about errors in readahead. */
8574 if (wc->stage == DROP_REFERENCE) {
8578 if (wc->level == 1 &&
8579 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8581 if (!wc->update_ref ||
8582 generation <= root->root_key.offset)
8584 btrfs_node_key_to_cpu(eb, &key, slot);
8585 ret = btrfs_comp_cpu_keys(&key,
8586 &wc->update_progress);
8590 if (wc->level == 1 &&
8591 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8595 readahead_tree_block(fs_info, bytenr);
8598 wc->reada_slot = slot;
8602 * helper to process tree block while walking down the tree.
8604 * when wc->stage == UPDATE_BACKREF, this function updates
8605 * back refs for pointers in the block.
8607 * NOTE: return value 1 means we should stop walking down.
8609 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8610 struct btrfs_root *root,
8611 struct btrfs_path *path,
8612 struct walk_control *wc, int lookup_info)
8614 struct btrfs_fs_info *fs_info = root->fs_info;
8615 int level = wc->level;
8616 struct extent_buffer *eb = path->nodes[level];
8617 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8620 if (wc->stage == UPDATE_BACKREF &&
8621 btrfs_header_owner(eb) != root->root_key.objectid)
8625 * when reference count of tree block is 1, it won't increase
8626 * again. once full backref flag is set, we never clear it.
8629 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8630 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8631 BUG_ON(!path->locks[level]);
8632 ret = btrfs_lookup_extent_info(trans, fs_info,
8633 eb->start, level, 1,
8636 BUG_ON(ret == -ENOMEM);
8639 BUG_ON(wc->refs[level] == 0);
8642 if (wc->stage == DROP_REFERENCE) {
8643 if (wc->refs[level] > 1)
8646 if (path->locks[level] && !wc->keep_locks) {
8647 btrfs_tree_unlock_rw(eb, path->locks[level]);
8648 path->locks[level] = 0;
8653 /* wc->stage == UPDATE_BACKREF */
8654 if (!(wc->flags[level] & flag)) {
8655 BUG_ON(!path->locks[level]);
8656 ret = btrfs_inc_ref(trans, root, eb, 1);
8657 BUG_ON(ret); /* -ENOMEM */
8658 ret = btrfs_dec_ref(trans, root, eb, 0);
8659 BUG_ON(ret); /* -ENOMEM */
8660 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8662 btrfs_header_level(eb), 0);
8663 BUG_ON(ret); /* -ENOMEM */
8664 wc->flags[level] |= flag;
8668 * the block is shared by multiple trees, so it's not good to
8669 * keep the tree lock
8671 if (path->locks[level] && level > 0) {
8672 btrfs_tree_unlock_rw(eb, path->locks[level]);
8673 path->locks[level] = 0;
8679 * helper to process tree block pointer.
8681 * when wc->stage == DROP_REFERENCE, this function checks
8682 * reference count of the block pointed to. if the block
8683 * is shared and we need update back refs for the subtree
8684 * rooted at the block, this function changes wc->stage to
8685 * UPDATE_BACKREF. if the block is shared and there is no
8686 * need to update back, this function drops the reference
8689 * NOTE: return value 1 means we should stop walking down.
8691 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8692 struct btrfs_root *root,
8693 struct btrfs_path *path,
8694 struct walk_control *wc, int *lookup_info)
8696 struct btrfs_fs_info *fs_info = root->fs_info;
8701 struct btrfs_key key;
8702 struct btrfs_key first_key;
8703 struct extent_buffer *next;
8704 int level = wc->level;
8707 bool need_account = false;
8709 generation = btrfs_node_ptr_generation(path->nodes[level],
8710 path->slots[level]);
8712 * if the lower level block was created before the snapshot
8713 * was created, we know there is no need to update back refs
8716 if (wc->stage == UPDATE_BACKREF &&
8717 generation <= root->root_key.offset) {
8722 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8723 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8724 path->slots[level]);
8725 blocksize = fs_info->nodesize;
8727 next = find_extent_buffer(fs_info, bytenr);
8729 next = btrfs_find_create_tree_block(fs_info, bytenr);
8731 return PTR_ERR(next);
8733 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8737 btrfs_tree_lock(next);
8738 btrfs_set_lock_blocking(next);
8740 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8741 &wc->refs[level - 1],
8742 &wc->flags[level - 1]);
8746 if (unlikely(wc->refs[level - 1] == 0)) {
8747 btrfs_err(fs_info, "Missing references.");
8753 if (wc->stage == DROP_REFERENCE) {
8754 if (wc->refs[level - 1] > 1) {
8755 need_account = true;
8757 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8760 if (!wc->update_ref ||
8761 generation <= root->root_key.offset)
8764 btrfs_node_key_to_cpu(path->nodes[level], &key,
8765 path->slots[level]);
8766 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8770 wc->stage = UPDATE_BACKREF;
8771 wc->shared_level = level - 1;
8775 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8779 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8780 btrfs_tree_unlock(next);
8781 free_extent_buffer(next);
8787 if (reada && level == 1)
8788 reada_walk_down(trans, root, wc, path);
8789 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8792 return PTR_ERR(next);
8793 } else if (!extent_buffer_uptodate(next)) {
8794 free_extent_buffer(next);
8797 btrfs_tree_lock(next);
8798 btrfs_set_lock_blocking(next);
8802 ASSERT(level == btrfs_header_level(next));
8803 if (level != btrfs_header_level(next)) {
8804 btrfs_err(root->fs_info, "mismatched level");
8808 path->nodes[level] = next;
8809 path->slots[level] = 0;
8810 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8816 wc->refs[level - 1] = 0;
8817 wc->flags[level - 1] = 0;
8818 if (wc->stage == DROP_REFERENCE) {
8819 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8820 parent = path->nodes[level]->start;
8822 ASSERT(root->root_key.objectid ==
8823 btrfs_header_owner(path->nodes[level]));
8824 if (root->root_key.objectid !=
8825 btrfs_header_owner(path->nodes[level])) {
8826 btrfs_err(root->fs_info,
8827 "mismatched block owner");
8835 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8836 generation, level - 1);
8838 btrfs_err_rl(fs_info,
8839 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8843 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8844 parent, root->root_key.objectid,
8854 btrfs_tree_unlock(next);
8855 free_extent_buffer(next);
8861 * helper to process tree block while walking up the tree.
8863 * when wc->stage == DROP_REFERENCE, this function drops
8864 * reference count on the block.
8866 * when wc->stage == UPDATE_BACKREF, this function changes
8867 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8868 * to UPDATE_BACKREF previously while processing the block.
8870 * NOTE: return value 1 means we should stop walking up.
8872 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8873 struct btrfs_root *root,
8874 struct btrfs_path *path,
8875 struct walk_control *wc)
8877 struct btrfs_fs_info *fs_info = root->fs_info;
8879 int level = wc->level;
8880 struct extent_buffer *eb = path->nodes[level];
8883 if (wc->stage == UPDATE_BACKREF) {
8884 BUG_ON(wc->shared_level < level);
8885 if (level < wc->shared_level)
8888 ret = find_next_key(path, level + 1, &wc->update_progress);
8892 wc->stage = DROP_REFERENCE;
8893 wc->shared_level = -1;
8894 path->slots[level] = 0;
8897 * check reference count again if the block isn't locked.
8898 * we should start walking down the tree again if reference
8901 if (!path->locks[level]) {
8903 btrfs_tree_lock(eb);
8904 btrfs_set_lock_blocking(eb);
8905 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8907 ret = btrfs_lookup_extent_info(trans, fs_info,
8908 eb->start, level, 1,
8912 btrfs_tree_unlock_rw(eb, path->locks[level]);
8913 path->locks[level] = 0;
8916 BUG_ON(wc->refs[level] == 0);
8917 if (wc->refs[level] == 1) {
8918 btrfs_tree_unlock_rw(eb, path->locks[level]);
8919 path->locks[level] = 0;
8925 /* wc->stage == DROP_REFERENCE */
8926 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8928 if (wc->refs[level] == 1) {
8930 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8931 ret = btrfs_dec_ref(trans, root, eb, 1);
8933 ret = btrfs_dec_ref(trans, root, eb, 0);
8934 BUG_ON(ret); /* -ENOMEM */
8935 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8937 btrfs_err_rl(fs_info,
8938 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8942 /* make block locked assertion in clean_tree_block happy */
8943 if (!path->locks[level] &&
8944 btrfs_header_generation(eb) == trans->transid) {
8945 btrfs_tree_lock(eb);
8946 btrfs_set_lock_blocking(eb);
8947 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8949 clean_tree_block(fs_info, eb);
8952 if (eb == root->node) {
8953 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8956 BUG_ON(root->root_key.objectid !=
8957 btrfs_header_owner(eb));
8959 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8960 parent = path->nodes[level + 1]->start;
8962 BUG_ON(root->root_key.objectid !=
8963 btrfs_header_owner(path->nodes[level + 1]));
8966 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8968 wc->refs[level] = 0;
8969 wc->flags[level] = 0;
8973 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8974 struct btrfs_root *root,
8975 struct btrfs_path *path,
8976 struct walk_control *wc)
8978 int level = wc->level;
8979 int lookup_info = 1;
8982 while (level >= 0) {
8983 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8990 if (path->slots[level] >=
8991 btrfs_header_nritems(path->nodes[level]))
8994 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8996 path->slots[level]++;
9005 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9006 struct btrfs_root *root,
9007 struct btrfs_path *path,
9008 struct walk_control *wc, int max_level)
9010 int level = wc->level;
9013 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9014 while (level < max_level && path->nodes[level]) {
9016 if (path->slots[level] + 1 <
9017 btrfs_header_nritems(path->nodes[level])) {
9018 path->slots[level]++;
9021 ret = walk_up_proc(trans, root, path, wc);
9025 if (path->locks[level]) {
9026 btrfs_tree_unlock_rw(path->nodes[level],
9027 path->locks[level]);
9028 path->locks[level] = 0;
9030 free_extent_buffer(path->nodes[level]);
9031 path->nodes[level] = NULL;
9039 * drop a subvolume tree.
9041 * this function traverses the tree freeing any blocks that only
9042 * referenced by the tree.
9044 * when a shared tree block is found. this function decreases its
9045 * reference count by one. if update_ref is true, this function
9046 * also make sure backrefs for the shared block and all lower level
9047 * blocks are properly updated.
9049 * If called with for_reloc == 0, may exit early with -EAGAIN
9051 int btrfs_drop_snapshot(struct btrfs_root *root,
9052 struct btrfs_block_rsv *block_rsv, int update_ref,
9055 struct btrfs_fs_info *fs_info = root->fs_info;
9056 struct btrfs_path *path;
9057 struct btrfs_trans_handle *trans;
9058 struct btrfs_root *tree_root = fs_info->tree_root;
9059 struct btrfs_root_item *root_item = &root->root_item;
9060 struct walk_control *wc;
9061 struct btrfs_key key;
9065 bool root_dropped = false;
9067 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9069 path = btrfs_alloc_path();
9075 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9077 btrfs_free_path(path);
9082 trans = btrfs_start_transaction(tree_root, 0);
9083 if (IS_ERR(trans)) {
9084 err = PTR_ERR(trans);
9089 trans->block_rsv = block_rsv;
9091 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9092 level = btrfs_header_level(root->node);
9093 path->nodes[level] = btrfs_lock_root_node(root);
9094 btrfs_set_lock_blocking(path->nodes[level]);
9095 path->slots[level] = 0;
9096 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9097 memset(&wc->update_progress, 0,
9098 sizeof(wc->update_progress));
9100 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9101 memcpy(&wc->update_progress, &key,
9102 sizeof(wc->update_progress));
9104 level = root_item->drop_level;
9106 path->lowest_level = level;
9107 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9108 path->lowest_level = 0;
9116 * unlock our path, this is safe because only this
9117 * function is allowed to delete this snapshot
9119 btrfs_unlock_up_safe(path, 0);
9121 level = btrfs_header_level(root->node);
9123 btrfs_tree_lock(path->nodes[level]);
9124 btrfs_set_lock_blocking(path->nodes[level]);
9125 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9127 ret = btrfs_lookup_extent_info(trans, fs_info,
9128 path->nodes[level]->start,
9129 level, 1, &wc->refs[level],
9135 BUG_ON(wc->refs[level] == 0);
9137 if (level == root_item->drop_level)
9140 btrfs_tree_unlock(path->nodes[level]);
9141 path->locks[level] = 0;
9142 WARN_ON(wc->refs[level] != 1);
9148 wc->shared_level = -1;
9149 wc->stage = DROP_REFERENCE;
9150 wc->update_ref = update_ref;
9152 wc->for_reloc = for_reloc;
9153 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9157 ret = walk_down_tree(trans, root, path, wc);
9163 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9170 BUG_ON(wc->stage != DROP_REFERENCE);
9174 if (wc->stage == DROP_REFERENCE) {
9176 btrfs_node_key(path->nodes[level],
9177 &root_item->drop_progress,
9178 path->slots[level]);
9179 root_item->drop_level = level;
9182 BUG_ON(wc->level == 0);
9183 if (btrfs_should_end_transaction(trans) ||
9184 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9185 ret = btrfs_update_root(trans, tree_root,
9189 btrfs_abort_transaction(trans, ret);
9194 btrfs_end_transaction_throttle(trans);
9195 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9196 btrfs_debug(fs_info,
9197 "drop snapshot early exit");
9202 trans = btrfs_start_transaction(tree_root, 0);
9203 if (IS_ERR(trans)) {
9204 err = PTR_ERR(trans);
9208 trans->block_rsv = block_rsv;
9211 btrfs_release_path(path);
9215 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9217 btrfs_abort_transaction(trans, ret);
9222 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9223 ret = btrfs_find_root(tree_root, &root->root_key, path,
9226 btrfs_abort_transaction(trans, ret);
9229 } else if (ret > 0) {
9230 /* if we fail to delete the orphan item this time
9231 * around, it'll get picked up the next time.
9233 * The most common failure here is just -ENOENT.
9235 btrfs_del_orphan_item(trans, tree_root,
9236 root->root_key.objectid);
9240 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9241 btrfs_add_dropped_root(trans, root);
9243 free_extent_buffer(root->node);
9244 free_extent_buffer(root->commit_root);
9245 btrfs_put_fs_root(root);
9247 root_dropped = true;
9249 btrfs_end_transaction_throttle(trans);
9252 btrfs_free_path(path);
9255 * So if we need to stop dropping the snapshot for whatever reason we
9256 * need to make sure to add it back to the dead root list so that we
9257 * keep trying to do the work later. This also cleans up roots if we
9258 * don't have it in the radix (like when we recover after a power fail
9259 * or unmount) so we don't leak memory.
9261 if (!for_reloc && !root_dropped)
9262 btrfs_add_dead_root(root);
9263 if (err && err != -EAGAIN)
9264 btrfs_handle_fs_error(fs_info, err, NULL);
9269 * drop subtree rooted at tree block 'node'.
9271 * NOTE: this function will unlock and release tree block 'node'
9272 * only used by relocation code
9274 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9275 struct btrfs_root *root,
9276 struct extent_buffer *node,
9277 struct extent_buffer *parent)
9279 struct btrfs_fs_info *fs_info = root->fs_info;
9280 struct btrfs_path *path;
9281 struct walk_control *wc;
9287 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9289 path = btrfs_alloc_path();
9293 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9295 btrfs_free_path(path);
9299 btrfs_assert_tree_locked(parent);
9300 parent_level = btrfs_header_level(parent);
9301 extent_buffer_get(parent);
9302 path->nodes[parent_level] = parent;
9303 path->slots[parent_level] = btrfs_header_nritems(parent);
9305 btrfs_assert_tree_locked(node);
9306 level = btrfs_header_level(node);
9307 path->nodes[level] = node;
9308 path->slots[level] = 0;
9309 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9311 wc->refs[parent_level] = 1;
9312 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9314 wc->shared_level = -1;
9315 wc->stage = DROP_REFERENCE;
9319 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9322 wret = walk_down_tree(trans, root, path, wc);
9328 wret = walk_up_tree(trans, root, path, wc, parent_level);
9336 btrfs_free_path(path);
9340 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9346 * if restripe for this chunk_type is on pick target profile and
9347 * return, otherwise do the usual balance
9349 stripped = get_restripe_target(fs_info, flags);
9351 return extended_to_chunk(stripped);
9353 num_devices = fs_info->fs_devices->rw_devices;
9355 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9356 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9357 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9359 if (num_devices == 1) {
9360 stripped |= BTRFS_BLOCK_GROUP_DUP;
9361 stripped = flags & ~stripped;
9363 /* turn raid0 into single device chunks */
9364 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9367 /* turn mirroring into duplication */
9368 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9369 BTRFS_BLOCK_GROUP_RAID10))
9370 return stripped | BTRFS_BLOCK_GROUP_DUP;
9372 /* they already had raid on here, just return */
9373 if (flags & stripped)
9376 stripped |= BTRFS_BLOCK_GROUP_DUP;
9377 stripped = flags & ~stripped;
9379 /* switch duplicated blocks with raid1 */
9380 if (flags & BTRFS_BLOCK_GROUP_DUP)
9381 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9383 /* this is drive concat, leave it alone */
9389 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9391 struct btrfs_space_info *sinfo = cache->space_info;
9393 u64 min_allocable_bytes;
9397 * We need some metadata space and system metadata space for
9398 * allocating chunks in some corner cases until we force to set
9399 * it to be readonly.
9402 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9404 min_allocable_bytes = SZ_1M;
9406 min_allocable_bytes = 0;
9408 spin_lock(&sinfo->lock);
9409 spin_lock(&cache->lock);
9417 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9418 cache->bytes_super - btrfs_block_group_used(&cache->item);
9420 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9421 min_allocable_bytes <= sinfo->total_bytes) {
9422 sinfo->bytes_readonly += num_bytes;
9424 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9428 spin_unlock(&cache->lock);
9429 spin_unlock(&sinfo->lock);
9433 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9434 struct btrfs_block_group_cache *cache)
9437 struct btrfs_trans_handle *trans;
9442 trans = btrfs_join_transaction(fs_info->extent_root);
9444 return PTR_ERR(trans);
9447 * we're not allowed to set block groups readonly after the dirty
9448 * block groups cache has started writing. If it already started,
9449 * back off and let this transaction commit
9451 mutex_lock(&fs_info->ro_block_group_mutex);
9452 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9453 u64 transid = trans->transid;
9455 mutex_unlock(&fs_info->ro_block_group_mutex);
9456 btrfs_end_transaction(trans);
9458 ret = btrfs_wait_for_commit(fs_info, transid);
9465 * if we are changing raid levels, try to allocate a corresponding
9466 * block group with the new raid level.
9468 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9469 if (alloc_flags != cache->flags) {
9470 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9473 * ENOSPC is allowed here, we may have enough space
9474 * already allocated at the new raid level to
9483 ret = inc_block_group_ro(cache, 0);
9486 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9487 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9491 ret = inc_block_group_ro(cache, 0);
9493 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9494 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9495 mutex_lock(&fs_info->chunk_mutex);
9496 check_system_chunk(trans, fs_info, alloc_flags);
9497 mutex_unlock(&fs_info->chunk_mutex);
9499 mutex_unlock(&fs_info->ro_block_group_mutex);
9501 btrfs_end_transaction(trans);
9505 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9506 struct btrfs_fs_info *fs_info, u64 type)
9508 u64 alloc_flags = get_alloc_profile(fs_info, type);
9510 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9514 * helper to account the unused space of all the readonly block group in the
9515 * space_info. takes mirrors into account.
9517 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9519 struct btrfs_block_group_cache *block_group;
9523 /* It's df, we don't care if it's racy */
9524 if (list_empty(&sinfo->ro_bgs))
9527 spin_lock(&sinfo->lock);
9528 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9529 spin_lock(&block_group->lock);
9531 if (!block_group->ro) {
9532 spin_unlock(&block_group->lock);
9536 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9537 BTRFS_BLOCK_GROUP_RAID10 |
9538 BTRFS_BLOCK_GROUP_DUP))
9543 free_bytes += (block_group->key.offset -
9544 btrfs_block_group_used(&block_group->item)) *
9547 spin_unlock(&block_group->lock);
9549 spin_unlock(&sinfo->lock);
9554 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9556 struct btrfs_space_info *sinfo = cache->space_info;
9561 spin_lock(&sinfo->lock);
9562 spin_lock(&cache->lock);
9564 num_bytes = cache->key.offset - cache->reserved -
9565 cache->pinned - cache->bytes_super -
9566 btrfs_block_group_used(&cache->item);
9567 sinfo->bytes_readonly -= num_bytes;
9568 list_del_init(&cache->ro_list);
9570 spin_unlock(&cache->lock);
9571 spin_unlock(&sinfo->lock);
9575 * checks to see if its even possible to relocate this block group.
9577 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9578 * ok to go ahead and try.
9580 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9582 struct btrfs_root *root = fs_info->extent_root;
9583 struct btrfs_block_group_cache *block_group;
9584 struct btrfs_space_info *space_info;
9585 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9586 struct btrfs_device *device;
9587 struct btrfs_trans_handle *trans;
9597 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9599 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9601 /* odd, couldn't find the block group, leave it alone */
9605 "can't find block group for bytenr %llu",
9610 min_free = btrfs_block_group_used(&block_group->item);
9612 /* no bytes used, we're good */
9616 space_info = block_group->space_info;
9617 spin_lock(&space_info->lock);
9619 full = space_info->full;
9622 * if this is the last block group we have in this space, we can't
9623 * relocate it unless we're able to allocate a new chunk below.
9625 * Otherwise, we need to make sure we have room in the space to handle
9626 * all of the extents from this block group. If we can, we're good
9628 if ((space_info->total_bytes != block_group->key.offset) &&
9629 (btrfs_space_info_used(space_info, false) + min_free <
9630 space_info->total_bytes)) {
9631 spin_unlock(&space_info->lock);
9634 spin_unlock(&space_info->lock);
9637 * ok we don't have enough space, but maybe we have free space on our
9638 * devices to allocate new chunks for relocation, so loop through our
9639 * alloc devices and guess if we have enough space. if this block
9640 * group is going to be restriped, run checks against the target
9641 * profile instead of the current one.
9653 target = get_restripe_target(fs_info, block_group->flags);
9655 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9658 * this is just a balance, so if we were marked as full
9659 * we know there is no space for a new chunk
9664 "no space to alloc new chunk for block group %llu",
9665 block_group->key.objectid);
9669 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9672 if (index == BTRFS_RAID_RAID10) {
9676 } else if (index == BTRFS_RAID_RAID1) {
9678 } else if (index == BTRFS_RAID_DUP) {
9681 } else if (index == BTRFS_RAID_RAID0) {
9682 dev_min = fs_devices->rw_devices;
9683 min_free = div64_u64(min_free, dev_min);
9686 /* We need to do this so that we can look at pending chunks */
9687 trans = btrfs_join_transaction(root);
9688 if (IS_ERR(trans)) {
9689 ret = PTR_ERR(trans);
9693 mutex_lock(&fs_info->chunk_mutex);
9694 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9698 * check to make sure we can actually find a chunk with enough
9699 * space to fit our block group in.
9701 if (device->total_bytes > device->bytes_used + min_free &&
9702 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9703 ret = find_free_dev_extent(trans, device, min_free,
9708 if (dev_nr >= dev_min)
9714 if (debug && ret == -1)
9716 "no space to allocate a new chunk for block group %llu",
9717 block_group->key.objectid);
9718 mutex_unlock(&fs_info->chunk_mutex);
9719 btrfs_end_transaction(trans);
9721 btrfs_put_block_group(block_group);
9725 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9726 struct btrfs_path *path,
9727 struct btrfs_key *key)
9729 struct btrfs_root *root = fs_info->extent_root;
9731 struct btrfs_key found_key;
9732 struct extent_buffer *leaf;
9735 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9740 slot = path->slots[0];
9741 leaf = path->nodes[0];
9742 if (slot >= btrfs_header_nritems(leaf)) {
9743 ret = btrfs_next_leaf(root, path);
9750 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9752 if (found_key.objectid >= key->objectid &&
9753 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9754 struct extent_map_tree *em_tree;
9755 struct extent_map *em;
9757 em_tree = &root->fs_info->mapping_tree.map_tree;
9758 read_lock(&em_tree->lock);
9759 em = lookup_extent_mapping(em_tree, found_key.objectid,
9761 read_unlock(&em_tree->lock);
9764 "logical %llu len %llu found bg but no related chunk",
9765 found_key.objectid, found_key.offset);
9770 free_extent_map(em);
9779 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9781 struct btrfs_block_group_cache *block_group;
9785 struct inode *inode;
9787 block_group = btrfs_lookup_first_block_group(info, last);
9788 while (block_group) {
9789 spin_lock(&block_group->lock);
9790 if (block_group->iref)
9792 spin_unlock(&block_group->lock);
9793 block_group = next_block_group(info, block_group);
9802 inode = block_group->inode;
9803 block_group->iref = 0;
9804 block_group->inode = NULL;
9805 spin_unlock(&block_group->lock);
9806 ASSERT(block_group->io_ctl.inode == NULL);
9808 last = block_group->key.objectid + block_group->key.offset;
9809 btrfs_put_block_group(block_group);
9814 * Must be called only after stopping all workers, since we could have block
9815 * group caching kthreads running, and therefore they could race with us if we
9816 * freed the block groups before stopping them.
9818 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9820 struct btrfs_block_group_cache *block_group;
9821 struct btrfs_space_info *space_info;
9822 struct btrfs_caching_control *caching_ctl;
9825 down_write(&info->commit_root_sem);
9826 while (!list_empty(&info->caching_block_groups)) {
9827 caching_ctl = list_entry(info->caching_block_groups.next,
9828 struct btrfs_caching_control, list);
9829 list_del(&caching_ctl->list);
9830 put_caching_control(caching_ctl);
9832 up_write(&info->commit_root_sem);
9834 spin_lock(&info->unused_bgs_lock);
9835 while (!list_empty(&info->unused_bgs)) {
9836 block_group = list_first_entry(&info->unused_bgs,
9837 struct btrfs_block_group_cache,
9839 list_del_init(&block_group->bg_list);
9840 btrfs_put_block_group(block_group);
9842 spin_unlock(&info->unused_bgs_lock);
9844 spin_lock(&info->block_group_cache_lock);
9845 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9846 block_group = rb_entry(n, struct btrfs_block_group_cache,
9848 rb_erase(&block_group->cache_node,
9849 &info->block_group_cache_tree);
9850 RB_CLEAR_NODE(&block_group->cache_node);
9851 spin_unlock(&info->block_group_cache_lock);
9853 down_write(&block_group->space_info->groups_sem);
9854 list_del(&block_group->list);
9855 up_write(&block_group->space_info->groups_sem);
9858 * We haven't cached this block group, which means we could
9859 * possibly have excluded extents on this block group.
9861 if (block_group->cached == BTRFS_CACHE_NO ||
9862 block_group->cached == BTRFS_CACHE_ERROR)
9863 free_excluded_extents(info, block_group);
9865 btrfs_remove_free_space_cache(block_group);
9866 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9867 ASSERT(list_empty(&block_group->dirty_list));
9868 ASSERT(list_empty(&block_group->io_list));
9869 ASSERT(list_empty(&block_group->bg_list));
9870 ASSERT(atomic_read(&block_group->count) == 1);
9871 btrfs_put_block_group(block_group);
9873 spin_lock(&info->block_group_cache_lock);
9875 spin_unlock(&info->block_group_cache_lock);
9877 /* now that all the block groups are freed, go through and
9878 * free all the space_info structs. This is only called during
9879 * the final stages of unmount, and so we know nobody is
9880 * using them. We call synchronize_rcu() once before we start,
9881 * just to be on the safe side.
9885 release_global_block_rsv(info);
9887 while (!list_empty(&info->space_info)) {
9890 space_info = list_entry(info->space_info.next,
9891 struct btrfs_space_info,
9895 * Do not hide this behind enospc_debug, this is actually
9896 * important and indicates a real bug if this happens.
9898 if (WARN_ON(space_info->bytes_pinned > 0 ||
9899 space_info->bytes_reserved > 0 ||
9900 space_info->bytes_may_use > 0))
9901 dump_space_info(info, space_info, 0, 0);
9902 list_del(&space_info->list);
9903 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9904 struct kobject *kobj;
9905 kobj = space_info->block_group_kobjs[i];
9906 space_info->block_group_kobjs[i] = NULL;
9912 kobject_del(&space_info->kobj);
9913 kobject_put(&space_info->kobj);
9918 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9919 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9921 struct btrfs_space_info *space_info;
9922 struct raid_kobject *rkobj;
9927 spin_lock(&fs_info->pending_raid_kobjs_lock);
9928 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9929 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9931 list_for_each_entry(rkobj, &list, list) {
9932 space_info = __find_space_info(fs_info, rkobj->flags);
9933 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9935 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9936 "%s", get_raid_name(index));
9938 kobject_put(&rkobj->kobj);
9944 "failed to add kobject for block cache, ignoring");
9947 static void link_block_group(struct btrfs_block_group_cache *cache)
9949 struct btrfs_space_info *space_info = cache->space_info;
9950 struct btrfs_fs_info *fs_info = cache->fs_info;
9951 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9954 down_write(&space_info->groups_sem);
9955 if (list_empty(&space_info->block_groups[index]))
9957 list_add_tail(&cache->list, &space_info->block_groups[index]);
9958 up_write(&space_info->groups_sem);
9961 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9963 btrfs_warn(cache->fs_info,
9964 "couldn't alloc memory for raid level kobject");
9967 rkobj->flags = cache->flags;
9968 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9970 spin_lock(&fs_info->pending_raid_kobjs_lock);
9971 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9972 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9973 space_info->block_group_kobjs[index] = &rkobj->kobj;
9977 static struct btrfs_block_group_cache *
9978 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9979 u64 start, u64 size)
9981 struct btrfs_block_group_cache *cache;
9983 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9987 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9989 if (!cache->free_space_ctl) {
9994 cache->key.objectid = start;
9995 cache->key.offset = size;
9996 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9998 cache->fs_info = fs_info;
9999 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10000 set_free_space_tree_thresholds(cache);
10002 atomic_set(&cache->count, 1);
10003 spin_lock_init(&cache->lock);
10004 init_rwsem(&cache->data_rwsem);
10005 INIT_LIST_HEAD(&cache->list);
10006 INIT_LIST_HEAD(&cache->cluster_list);
10007 INIT_LIST_HEAD(&cache->bg_list);
10008 INIT_LIST_HEAD(&cache->ro_list);
10009 INIT_LIST_HEAD(&cache->dirty_list);
10010 INIT_LIST_HEAD(&cache->io_list);
10011 btrfs_init_free_space_ctl(cache);
10012 atomic_set(&cache->trimming, 0);
10013 mutex_init(&cache->free_space_lock);
10014 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10019 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10021 struct btrfs_path *path;
10023 struct btrfs_block_group_cache *cache;
10024 struct btrfs_space_info *space_info;
10025 struct btrfs_key key;
10026 struct btrfs_key found_key;
10027 struct extent_buffer *leaf;
10028 int need_clear = 0;
10033 feature = btrfs_super_incompat_flags(info->super_copy);
10034 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10038 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10039 path = btrfs_alloc_path();
10042 path->reada = READA_FORWARD;
10044 cache_gen = btrfs_super_cache_generation(info->super_copy);
10045 if (btrfs_test_opt(info, SPACE_CACHE) &&
10046 btrfs_super_generation(info->super_copy) != cache_gen)
10048 if (btrfs_test_opt(info, CLEAR_CACHE))
10052 ret = find_first_block_group(info, path, &key);
10058 leaf = path->nodes[0];
10059 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10061 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10070 * When we mount with old space cache, we need to
10071 * set BTRFS_DC_CLEAR and set dirty flag.
10073 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10074 * truncate the old free space cache inode and
10076 * b) Setting 'dirty flag' makes sure that we flush
10077 * the new space cache info onto disk.
10079 if (btrfs_test_opt(info, SPACE_CACHE))
10080 cache->disk_cache_state = BTRFS_DC_CLEAR;
10083 read_extent_buffer(leaf, &cache->item,
10084 btrfs_item_ptr_offset(leaf, path->slots[0]),
10085 sizeof(cache->item));
10086 cache->flags = btrfs_block_group_flags(&cache->item);
10088 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10089 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10091 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10092 cache->key.objectid);
10097 key.objectid = found_key.objectid + found_key.offset;
10098 btrfs_release_path(path);
10101 * We need to exclude the super stripes now so that the space
10102 * info has super bytes accounted for, otherwise we'll think
10103 * we have more space than we actually do.
10105 ret = exclude_super_stripes(info, cache);
10108 * We may have excluded something, so call this just in
10111 free_excluded_extents(info, cache);
10112 btrfs_put_block_group(cache);
10117 * check for two cases, either we are full, and therefore
10118 * don't need to bother with the caching work since we won't
10119 * find any space, or we are empty, and we can just add all
10120 * the space in and be done with it. This saves us _alot_ of
10121 * time, particularly in the full case.
10123 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10124 cache->last_byte_to_unpin = (u64)-1;
10125 cache->cached = BTRFS_CACHE_FINISHED;
10126 free_excluded_extents(info, cache);
10127 } else if (btrfs_block_group_used(&cache->item) == 0) {
10128 cache->last_byte_to_unpin = (u64)-1;
10129 cache->cached = BTRFS_CACHE_FINISHED;
10130 add_new_free_space(cache, info,
10131 found_key.objectid,
10132 found_key.objectid +
10134 free_excluded_extents(info, cache);
10137 ret = btrfs_add_block_group_cache(info, cache);
10139 btrfs_remove_free_space_cache(cache);
10140 btrfs_put_block_group(cache);
10144 trace_btrfs_add_block_group(info, cache, 0);
10145 update_space_info(info, cache->flags, found_key.offset,
10146 btrfs_block_group_used(&cache->item),
10147 cache->bytes_super, &space_info);
10149 cache->space_info = space_info;
10151 link_block_group(cache);
10153 set_avail_alloc_bits(info, cache->flags);
10154 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10155 inc_block_group_ro(cache, 1);
10156 } else if (btrfs_block_group_used(&cache->item) == 0) {
10157 spin_lock(&info->unused_bgs_lock);
10158 /* Should always be true but just in case. */
10159 if (list_empty(&cache->bg_list)) {
10160 btrfs_get_block_group(cache);
10161 list_add_tail(&cache->bg_list,
10162 &info->unused_bgs);
10164 spin_unlock(&info->unused_bgs_lock);
10168 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10169 if (!(get_alloc_profile(info, space_info->flags) &
10170 (BTRFS_BLOCK_GROUP_RAID10 |
10171 BTRFS_BLOCK_GROUP_RAID1 |
10172 BTRFS_BLOCK_GROUP_RAID5 |
10173 BTRFS_BLOCK_GROUP_RAID6 |
10174 BTRFS_BLOCK_GROUP_DUP)))
10177 * avoid allocating from un-mirrored block group if there are
10178 * mirrored block groups.
10180 list_for_each_entry(cache,
10181 &space_info->block_groups[BTRFS_RAID_RAID0],
10183 inc_block_group_ro(cache, 1);
10184 list_for_each_entry(cache,
10185 &space_info->block_groups[BTRFS_RAID_SINGLE],
10187 inc_block_group_ro(cache, 1);
10190 btrfs_add_raid_kobjects(info);
10191 init_global_block_rsv(info);
10194 btrfs_free_path(path);
10198 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10200 struct btrfs_fs_info *fs_info = trans->fs_info;
10201 struct btrfs_block_group_cache *block_group, *tmp;
10202 struct btrfs_root *extent_root = fs_info->extent_root;
10203 struct btrfs_block_group_item item;
10204 struct btrfs_key key;
10206 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10208 trans->can_flush_pending_bgs = false;
10209 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10213 spin_lock(&block_group->lock);
10214 memcpy(&item, &block_group->item, sizeof(item));
10215 memcpy(&key, &block_group->key, sizeof(key));
10216 spin_unlock(&block_group->lock);
10218 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10221 btrfs_abort_transaction(trans, ret);
10222 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10225 btrfs_abort_transaction(trans, ret);
10226 add_block_group_free_space(trans, fs_info, block_group);
10227 /* already aborted the transaction if it failed. */
10229 list_del_init(&block_group->bg_list);
10231 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10234 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10235 struct btrfs_fs_info *fs_info, u64 bytes_used,
10236 u64 type, u64 chunk_offset, u64 size)
10238 struct btrfs_block_group_cache *cache;
10241 btrfs_set_log_full_commit(fs_info, trans);
10243 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10247 btrfs_set_block_group_used(&cache->item, bytes_used);
10248 btrfs_set_block_group_chunk_objectid(&cache->item,
10249 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10250 btrfs_set_block_group_flags(&cache->item, type);
10252 cache->flags = type;
10253 cache->last_byte_to_unpin = (u64)-1;
10254 cache->cached = BTRFS_CACHE_FINISHED;
10255 cache->needs_free_space = 1;
10256 ret = exclude_super_stripes(fs_info, cache);
10259 * We may have excluded something, so call this just in
10262 free_excluded_extents(fs_info, cache);
10263 btrfs_put_block_group(cache);
10267 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10269 free_excluded_extents(fs_info, cache);
10271 #ifdef CONFIG_BTRFS_DEBUG
10272 if (btrfs_should_fragment_free_space(cache)) {
10273 u64 new_bytes_used = size - bytes_used;
10275 bytes_used += new_bytes_used >> 1;
10276 fragment_free_space(cache);
10280 * Ensure the corresponding space_info object is created and
10281 * assigned to our block group. We want our bg to be added to the rbtree
10282 * with its ->space_info set.
10284 cache->space_info = __find_space_info(fs_info, cache->flags);
10285 ASSERT(cache->space_info);
10287 ret = btrfs_add_block_group_cache(fs_info, cache);
10289 btrfs_remove_free_space_cache(cache);
10290 btrfs_put_block_group(cache);
10295 * Now that our block group has its ->space_info set and is inserted in
10296 * the rbtree, update the space info's counters.
10298 trace_btrfs_add_block_group(fs_info, cache, 1);
10299 update_space_info(fs_info, cache->flags, size, bytes_used,
10300 cache->bytes_super, &cache->space_info);
10301 update_global_block_rsv(fs_info);
10303 link_block_group(cache);
10305 list_add_tail(&cache->bg_list, &trans->new_bgs);
10307 set_avail_alloc_bits(fs_info, type);
10311 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10313 u64 extra_flags = chunk_to_extended(flags) &
10314 BTRFS_EXTENDED_PROFILE_MASK;
10316 write_seqlock(&fs_info->profiles_lock);
10317 if (flags & BTRFS_BLOCK_GROUP_DATA)
10318 fs_info->avail_data_alloc_bits &= ~extra_flags;
10319 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10320 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10321 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10322 fs_info->avail_system_alloc_bits &= ~extra_flags;
10323 write_sequnlock(&fs_info->profiles_lock);
10326 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10327 struct btrfs_fs_info *fs_info, u64 group_start,
10328 struct extent_map *em)
10330 struct btrfs_root *root = fs_info->extent_root;
10331 struct btrfs_path *path;
10332 struct btrfs_block_group_cache *block_group;
10333 struct btrfs_free_cluster *cluster;
10334 struct btrfs_root *tree_root = fs_info->tree_root;
10335 struct btrfs_key key;
10336 struct inode *inode;
10337 struct kobject *kobj = NULL;
10341 struct btrfs_caching_control *caching_ctl = NULL;
10344 block_group = btrfs_lookup_block_group(fs_info, group_start);
10345 BUG_ON(!block_group);
10346 BUG_ON(!block_group->ro);
10349 * Free the reserved super bytes from this block group before
10352 free_excluded_extents(fs_info, block_group);
10353 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10354 block_group->key.offset);
10356 memcpy(&key, &block_group->key, sizeof(key));
10357 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10358 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10359 BTRFS_BLOCK_GROUP_RAID1 |
10360 BTRFS_BLOCK_GROUP_RAID10))
10365 /* make sure this block group isn't part of an allocation cluster */
10366 cluster = &fs_info->data_alloc_cluster;
10367 spin_lock(&cluster->refill_lock);
10368 btrfs_return_cluster_to_free_space(block_group, cluster);
10369 spin_unlock(&cluster->refill_lock);
10372 * make sure this block group isn't part of a metadata
10373 * allocation cluster
10375 cluster = &fs_info->meta_alloc_cluster;
10376 spin_lock(&cluster->refill_lock);
10377 btrfs_return_cluster_to_free_space(block_group, cluster);
10378 spin_unlock(&cluster->refill_lock);
10380 path = btrfs_alloc_path();
10387 * get the inode first so any iput calls done for the io_list
10388 * aren't the final iput (no unlinks allowed now)
10390 inode = lookup_free_space_inode(fs_info, block_group, path);
10392 mutex_lock(&trans->transaction->cache_write_mutex);
10394 * make sure our free spache cache IO is done before remove the
10397 spin_lock(&trans->transaction->dirty_bgs_lock);
10398 if (!list_empty(&block_group->io_list)) {
10399 list_del_init(&block_group->io_list);
10401 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10403 spin_unlock(&trans->transaction->dirty_bgs_lock);
10404 btrfs_wait_cache_io(trans, block_group, path);
10405 btrfs_put_block_group(block_group);
10406 spin_lock(&trans->transaction->dirty_bgs_lock);
10409 if (!list_empty(&block_group->dirty_list)) {
10410 list_del_init(&block_group->dirty_list);
10411 btrfs_put_block_group(block_group);
10413 spin_unlock(&trans->transaction->dirty_bgs_lock);
10414 mutex_unlock(&trans->transaction->cache_write_mutex);
10416 if (!IS_ERR(inode)) {
10417 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10419 btrfs_add_delayed_iput(inode);
10422 clear_nlink(inode);
10423 /* One for the block groups ref */
10424 spin_lock(&block_group->lock);
10425 if (block_group->iref) {
10426 block_group->iref = 0;
10427 block_group->inode = NULL;
10428 spin_unlock(&block_group->lock);
10431 spin_unlock(&block_group->lock);
10433 /* One for our lookup ref */
10434 btrfs_add_delayed_iput(inode);
10437 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10438 key.offset = block_group->key.objectid;
10441 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10445 btrfs_release_path(path);
10447 ret = btrfs_del_item(trans, tree_root, path);
10450 btrfs_release_path(path);
10453 spin_lock(&fs_info->block_group_cache_lock);
10454 rb_erase(&block_group->cache_node,
10455 &fs_info->block_group_cache_tree);
10456 RB_CLEAR_NODE(&block_group->cache_node);
10458 if (fs_info->first_logical_byte == block_group->key.objectid)
10459 fs_info->first_logical_byte = (u64)-1;
10460 spin_unlock(&fs_info->block_group_cache_lock);
10462 down_write(&block_group->space_info->groups_sem);
10464 * we must use list_del_init so people can check to see if they
10465 * are still on the list after taking the semaphore
10467 list_del_init(&block_group->list);
10468 if (list_empty(&block_group->space_info->block_groups[index])) {
10469 kobj = block_group->space_info->block_group_kobjs[index];
10470 block_group->space_info->block_group_kobjs[index] = NULL;
10471 clear_avail_alloc_bits(fs_info, block_group->flags);
10473 up_write(&block_group->space_info->groups_sem);
10479 if (block_group->has_caching_ctl)
10480 caching_ctl = get_caching_control(block_group);
10481 if (block_group->cached == BTRFS_CACHE_STARTED)
10482 wait_block_group_cache_done(block_group);
10483 if (block_group->has_caching_ctl) {
10484 down_write(&fs_info->commit_root_sem);
10485 if (!caching_ctl) {
10486 struct btrfs_caching_control *ctl;
10488 list_for_each_entry(ctl,
10489 &fs_info->caching_block_groups, list)
10490 if (ctl->block_group == block_group) {
10492 refcount_inc(&caching_ctl->count);
10497 list_del_init(&caching_ctl->list);
10498 up_write(&fs_info->commit_root_sem);
10500 /* Once for the caching bgs list and once for us. */
10501 put_caching_control(caching_ctl);
10502 put_caching_control(caching_ctl);
10506 spin_lock(&trans->transaction->dirty_bgs_lock);
10507 if (!list_empty(&block_group->dirty_list)) {
10510 if (!list_empty(&block_group->io_list)) {
10513 spin_unlock(&trans->transaction->dirty_bgs_lock);
10514 btrfs_remove_free_space_cache(block_group);
10516 spin_lock(&block_group->space_info->lock);
10517 list_del_init(&block_group->ro_list);
10519 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10520 WARN_ON(block_group->space_info->total_bytes
10521 < block_group->key.offset);
10522 WARN_ON(block_group->space_info->bytes_readonly
10523 < block_group->key.offset);
10524 WARN_ON(block_group->space_info->disk_total
10525 < block_group->key.offset * factor);
10527 block_group->space_info->total_bytes -= block_group->key.offset;
10528 block_group->space_info->bytes_readonly -= block_group->key.offset;
10529 block_group->space_info->disk_total -= block_group->key.offset * factor;
10531 spin_unlock(&block_group->space_info->lock);
10533 memcpy(&key, &block_group->key, sizeof(key));
10535 mutex_lock(&fs_info->chunk_mutex);
10536 if (!list_empty(&em->list)) {
10537 /* We're in the transaction->pending_chunks list. */
10538 free_extent_map(em);
10540 spin_lock(&block_group->lock);
10541 block_group->removed = 1;
10543 * At this point trimming can't start on this block group, because we
10544 * removed the block group from the tree fs_info->block_group_cache_tree
10545 * so no one can't find it anymore and even if someone already got this
10546 * block group before we removed it from the rbtree, they have already
10547 * incremented block_group->trimming - if they didn't, they won't find
10548 * any free space entries because we already removed them all when we
10549 * called btrfs_remove_free_space_cache().
10551 * And we must not remove the extent map from the fs_info->mapping_tree
10552 * to prevent the same logical address range and physical device space
10553 * ranges from being reused for a new block group. This is because our
10554 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10555 * completely transactionless, so while it is trimming a range the
10556 * currently running transaction might finish and a new one start,
10557 * allowing for new block groups to be created that can reuse the same
10558 * physical device locations unless we take this special care.
10560 * There may also be an implicit trim operation if the file system
10561 * is mounted with -odiscard. The same protections must remain
10562 * in place until the extents have been discarded completely when
10563 * the transaction commit has completed.
10565 remove_em = (atomic_read(&block_group->trimming) == 0);
10567 * Make sure a trimmer task always sees the em in the pinned_chunks list
10568 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10569 * before checking block_group->removed).
10573 * Our em might be in trans->transaction->pending_chunks which
10574 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10575 * and so is the fs_info->pinned_chunks list.
10577 * So at this point we must be holding the chunk_mutex to avoid
10578 * any races with chunk allocation (more specifically at
10579 * volumes.c:contains_pending_extent()), to ensure it always
10580 * sees the em, either in the pending_chunks list or in the
10581 * pinned_chunks list.
10583 list_move_tail(&em->list, &fs_info->pinned_chunks);
10585 spin_unlock(&block_group->lock);
10588 struct extent_map_tree *em_tree;
10590 em_tree = &fs_info->mapping_tree.map_tree;
10591 write_lock(&em_tree->lock);
10593 * The em might be in the pending_chunks list, so make sure the
10594 * chunk mutex is locked, since remove_extent_mapping() will
10595 * delete us from that list.
10597 remove_extent_mapping(em_tree, em);
10598 write_unlock(&em_tree->lock);
10599 /* once for the tree */
10600 free_extent_map(em);
10603 mutex_unlock(&fs_info->chunk_mutex);
10605 ret = remove_block_group_free_space(trans, fs_info, block_group);
10609 btrfs_put_block_group(block_group);
10610 btrfs_put_block_group(block_group);
10612 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10618 ret = btrfs_del_item(trans, root, path);
10620 btrfs_free_path(path);
10624 struct btrfs_trans_handle *
10625 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10626 const u64 chunk_offset)
10628 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10629 struct extent_map *em;
10630 struct map_lookup *map;
10631 unsigned int num_items;
10633 read_lock(&em_tree->lock);
10634 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10635 read_unlock(&em_tree->lock);
10636 ASSERT(em && em->start == chunk_offset);
10639 * We need to reserve 3 + N units from the metadata space info in order
10640 * to remove a block group (done at btrfs_remove_chunk() and at
10641 * btrfs_remove_block_group()), which are used for:
10643 * 1 unit for adding the free space inode's orphan (located in the tree
10645 * 1 unit for deleting the block group item (located in the extent
10647 * 1 unit for deleting the free space item (located in tree of tree
10649 * N units for deleting N device extent items corresponding to each
10650 * stripe (located in the device tree).
10652 * In order to remove a block group we also need to reserve units in the
10653 * system space info in order to update the chunk tree (update one or
10654 * more device items and remove one chunk item), but this is done at
10655 * btrfs_remove_chunk() through a call to check_system_chunk().
10657 map = em->map_lookup;
10658 num_items = 3 + map->num_stripes;
10659 free_extent_map(em);
10661 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10666 * Process the unused_bgs list and remove any that don't have any allocated
10667 * space inside of them.
10669 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10671 struct btrfs_block_group_cache *block_group;
10672 struct btrfs_space_info *space_info;
10673 struct btrfs_trans_handle *trans;
10676 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10679 spin_lock(&fs_info->unused_bgs_lock);
10680 while (!list_empty(&fs_info->unused_bgs)) {
10684 block_group = list_first_entry(&fs_info->unused_bgs,
10685 struct btrfs_block_group_cache,
10687 list_del_init(&block_group->bg_list);
10689 space_info = block_group->space_info;
10691 if (ret || btrfs_mixed_space_info(space_info)) {
10692 btrfs_put_block_group(block_group);
10695 spin_unlock(&fs_info->unused_bgs_lock);
10697 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10699 /* Don't want to race with allocators so take the groups_sem */
10700 down_write(&space_info->groups_sem);
10701 spin_lock(&block_group->lock);
10702 if (block_group->reserved ||
10703 btrfs_block_group_used(&block_group->item) ||
10705 list_is_singular(&block_group->list)) {
10707 * We want to bail if we made new allocations or have
10708 * outstanding allocations in this block group. We do
10709 * the ro check in case balance is currently acting on
10710 * this block group.
10712 spin_unlock(&block_group->lock);
10713 up_write(&space_info->groups_sem);
10716 spin_unlock(&block_group->lock);
10718 /* We don't want to force the issue, only flip if it's ok. */
10719 ret = inc_block_group_ro(block_group, 0);
10720 up_write(&space_info->groups_sem);
10727 * Want to do this before we do anything else so we can recover
10728 * properly if we fail to join the transaction.
10730 trans = btrfs_start_trans_remove_block_group(fs_info,
10731 block_group->key.objectid);
10732 if (IS_ERR(trans)) {
10733 btrfs_dec_block_group_ro(block_group);
10734 ret = PTR_ERR(trans);
10739 * We could have pending pinned extents for this block group,
10740 * just delete them, we don't care about them anymore.
10742 start = block_group->key.objectid;
10743 end = start + block_group->key.offset - 1;
10745 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10746 * btrfs_finish_extent_commit(). If we are at transaction N,
10747 * another task might be running finish_extent_commit() for the
10748 * previous transaction N - 1, and have seen a range belonging
10749 * to the block group in freed_extents[] before we were able to
10750 * clear the whole block group range from freed_extents[]. This
10751 * means that task can lookup for the block group after we
10752 * unpinned it from freed_extents[] and removed it, leading to
10753 * a BUG_ON() at btrfs_unpin_extent_range().
10755 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10756 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10759 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10760 btrfs_dec_block_group_ro(block_group);
10763 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10766 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10767 btrfs_dec_block_group_ro(block_group);
10770 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10772 /* Reset pinned so btrfs_put_block_group doesn't complain */
10773 spin_lock(&space_info->lock);
10774 spin_lock(&block_group->lock);
10776 space_info->bytes_pinned -= block_group->pinned;
10777 space_info->bytes_readonly += block_group->pinned;
10778 percpu_counter_add(&space_info->total_bytes_pinned,
10779 -block_group->pinned);
10780 block_group->pinned = 0;
10782 spin_unlock(&block_group->lock);
10783 spin_unlock(&space_info->lock);
10785 /* DISCARD can flip during remount */
10786 trimming = btrfs_test_opt(fs_info, DISCARD);
10788 /* Implicit trim during transaction commit. */
10790 btrfs_get_block_group_trimming(block_group);
10793 * Btrfs_remove_chunk will abort the transaction if things go
10796 ret = btrfs_remove_chunk(trans, fs_info,
10797 block_group->key.objectid);
10801 btrfs_put_block_group_trimming(block_group);
10806 * If we're not mounted with -odiscard, we can just forget
10807 * about this block group. Otherwise we'll need to wait
10808 * until transaction commit to do the actual discard.
10811 spin_lock(&fs_info->unused_bgs_lock);
10813 * A concurrent scrub might have added us to the list
10814 * fs_info->unused_bgs, so use a list_move operation
10815 * to add the block group to the deleted_bgs list.
10817 list_move(&block_group->bg_list,
10818 &trans->transaction->deleted_bgs);
10819 spin_unlock(&fs_info->unused_bgs_lock);
10820 btrfs_get_block_group(block_group);
10823 btrfs_end_transaction(trans);
10825 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10826 btrfs_put_block_group(block_group);
10827 spin_lock(&fs_info->unused_bgs_lock);
10829 spin_unlock(&fs_info->unused_bgs_lock);
10832 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10834 struct btrfs_space_info *space_info;
10835 struct btrfs_super_block *disk_super;
10841 disk_super = fs_info->super_copy;
10842 if (!btrfs_super_root(disk_super))
10845 features = btrfs_super_incompat_flags(disk_super);
10846 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10849 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10850 ret = create_space_info(fs_info, flags, &space_info);
10855 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10856 ret = create_space_info(fs_info, flags, &space_info);
10858 flags = BTRFS_BLOCK_GROUP_METADATA;
10859 ret = create_space_info(fs_info, flags, &space_info);
10863 flags = BTRFS_BLOCK_GROUP_DATA;
10864 ret = create_space_info(fs_info, flags, &space_info);
10870 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10871 u64 start, u64 end)
10873 return unpin_extent_range(fs_info, start, end, false);
10877 * It used to be that old block groups would be left around forever.
10878 * Iterating over them would be enough to trim unused space. Since we
10879 * now automatically remove them, we also need to iterate over unallocated
10882 * We don't want a transaction for this since the discard may take a
10883 * substantial amount of time. We don't require that a transaction be
10884 * running, but we do need to take a running transaction into account
10885 * to ensure that we're not discarding chunks that were released in
10886 * the current transaction.
10888 * Holding the chunks lock will prevent other threads from allocating
10889 * or releasing chunks, but it won't prevent a running transaction
10890 * from committing and releasing the memory that the pending chunks
10891 * list head uses. For that, we need to take a reference to the
10894 static int btrfs_trim_free_extents(struct btrfs_device *device,
10895 u64 minlen, u64 *trimmed)
10897 u64 start = 0, len = 0;
10902 /* Not writeable = nothing to do. */
10903 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10906 /* No free space = nothing to do. */
10907 if (device->total_bytes <= device->bytes_used)
10913 struct btrfs_fs_info *fs_info = device->fs_info;
10914 struct btrfs_transaction *trans;
10917 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10921 down_read(&fs_info->commit_root_sem);
10923 spin_lock(&fs_info->trans_lock);
10924 trans = fs_info->running_transaction;
10926 refcount_inc(&trans->use_count);
10927 spin_unlock(&fs_info->trans_lock);
10929 ret = find_free_dev_extent_start(trans, device, minlen, start,
10932 btrfs_put_transaction(trans);
10935 up_read(&fs_info->commit_root_sem);
10936 mutex_unlock(&fs_info->chunk_mutex);
10937 if (ret == -ENOSPC)
10942 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10943 up_read(&fs_info->commit_root_sem);
10944 mutex_unlock(&fs_info->chunk_mutex);
10952 if (fatal_signal_pending(current)) {
10953 ret = -ERESTARTSYS;
10963 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10965 struct btrfs_block_group_cache *cache = NULL;
10966 struct btrfs_device *device;
10967 struct list_head *devices;
10972 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10976 * try to trim all FS space, our block group may start from non-zero.
10978 if (range->len == total_bytes)
10979 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10981 cache = btrfs_lookup_block_group(fs_info, range->start);
10984 if (cache->key.objectid >= (range->start + range->len)) {
10985 btrfs_put_block_group(cache);
10989 start = max(range->start, cache->key.objectid);
10990 end = min(range->start + range->len,
10991 cache->key.objectid + cache->key.offset);
10993 if (end - start >= range->minlen) {
10994 if (!block_group_cache_done(cache)) {
10995 ret = cache_block_group(cache, 0);
10997 btrfs_put_block_group(cache);
11000 ret = wait_block_group_cache_done(cache);
11002 btrfs_put_block_group(cache);
11006 ret = btrfs_trim_block_group(cache,
11012 trimmed += group_trimmed;
11014 btrfs_put_block_group(cache);
11019 cache = next_block_group(fs_info, cache);
11022 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11023 devices = &fs_info->fs_devices->alloc_list;
11024 list_for_each_entry(device, devices, dev_alloc_list) {
11025 ret = btrfs_trim_free_extents(device, range->minlen,
11030 trimmed += group_trimmed;
11032 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11034 range->len = trimmed;
11039 * btrfs_{start,end}_write_no_snapshotting() are similar to
11040 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11041 * data into the page cache through nocow before the subvolume is snapshoted,
11042 * but flush the data into disk after the snapshot creation, or to prevent
11043 * operations while snapshotting is ongoing and that cause the snapshot to be
11044 * inconsistent (writes followed by expanding truncates for example).
11046 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11048 percpu_counter_dec(&root->subv_writers->counter);
11050 * Make sure counter is updated before we wake up waiters.
11053 if (waitqueue_active(&root->subv_writers->wait))
11054 wake_up(&root->subv_writers->wait);
11057 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11059 if (atomic_read(&root->will_be_snapshotted))
11062 percpu_counter_inc(&root->subv_writers->counter);
11064 * Make sure counter is updated before we check for snapshot creation.
11067 if (atomic_read(&root->will_be_snapshotted)) {
11068 btrfs_end_write_no_snapshotting(root);
11074 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11079 ret = btrfs_start_write_no_snapshotting(root);
11082 wait_on_atomic_t(&root->will_be_snapshotted, atomic_t_wait,
11083 TASK_UNINTERRUPTIBLE);